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;
687 long long int consistency_rl;
690 struct table_dpif tables[N_TABLES];
692 /* Support for debugging async flow mods. */
693 struct list completions;
695 bool has_bundle_action; /* True when the first bundle action appears. */
696 struct netdev_stats stats; /* To account packets generated and consumed in
701 long long int stp_last_tick;
703 /* VLAN splinters. */
704 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
705 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
708 struct sset ports; /* Set of standard port names. */
709 struct sset ghost_ports; /* Ports with no datapath port. */
710 struct sset port_poll_set; /* Queued names for port_poll() reply. */
711 int port_poll_errno; /* Last errno for port_poll() reply. */
714 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
715 * for debugging the asynchronous flow_mod implementation.) */
718 /* All existing ofproto_dpif instances, indexed by ->up.name. */
719 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
721 static void ofproto_dpif_unixctl_init(void);
723 static struct ofproto_dpif *
724 ofproto_dpif_cast(const struct ofproto *ofproto)
726 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
727 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
730 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
732 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
734 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
735 const struct ofpbuf *,
736 const struct initial_vals *, struct ds *);
738 /* Packet processing. */
739 static void update_learning_table(struct ofproto_dpif *,
740 const struct flow *, int vlan,
743 #define FLOW_MISS_MAX_BATCH 50
744 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
746 /* Flow expiration. */
747 static int expire(struct dpif_backer *);
750 static void send_netflow_active_timeouts(struct ofproto_dpif *);
753 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
754 static size_t compose_sflow_action(const struct ofproto_dpif *,
755 struct ofpbuf *odp_actions,
756 const struct flow *, uint32_t odp_port);
757 static void add_mirror_actions(struct action_xlate_ctx *ctx,
758 const struct flow *flow);
759 /* Global variables. */
760 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
762 /* Initial mappings of port to bridge mappings. */
763 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
765 /* Factory functions. */
768 init(const struct shash *iface_hints)
770 struct shash_node *node;
772 /* Make a local copy, since we don't own 'iface_hints' elements. */
773 SHASH_FOR_EACH(node, iface_hints) {
774 const struct iface_hint *orig_hint = node->data;
775 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
777 new_hint->br_name = xstrdup(orig_hint->br_name);
778 new_hint->br_type = xstrdup(orig_hint->br_type);
779 new_hint->ofp_port = orig_hint->ofp_port;
781 shash_add(&init_ofp_ports, node->name, new_hint);
786 enumerate_types(struct sset *types)
788 dp_enumerate_types(types);
792 enumerate_names(const char *type, struct sset *names)
794 struct ofproto_dpif *ofproto;
797 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
798 if (strcmp(type, ofproto->up.type)) {
801 sset_add(names, ofproto->up.name);
808 del(const char *type, const char *name)
813 error = dpif_open(name, type, &dpif);
815 error = dpif_delete(dpif);
822 port_open_type(const char *datapath_type, const char *port_type)
824 return dpif_port_open_type(datapath_type, port_type);
827 /* Type functions. */
829 static struct ofproto_dpif *
830 lookup_ofproto_dpif_by_port_name(const char *name)
832 struct ofproto_dpif *ofproto;
834 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
835 if (sset_contains(&ofproto->ports, name)) {
844 type_run(const char *type)
846 struct dpif_backer *backer;
850 backer = shash_find_data(&all_dpif_backers, type);
852 /* This is not necessarily a problem, since backers are only
853 * created on demand. */
857 dpif_run(backer->dpif);
859 if (backer->need_revalidate
860 || !tag_set_is_empty(&backer->revalidate_set)) {
861 struct tag_set revalidate_set = backer->revalidate_set;
862 bool need_revalidate = backer->need_revalidate;
863 struct ofproto_dpif *ofproto;
864 struct simap_node *node;
865 struct simap tmp_backers;
867 /* Handle tunnel garbage collection. */
868 simap_init(&tmp_backers);
869 simap_swap(&backer->tnl_backers, &tmp_backers);
871 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
872 struct ofport_dpif *iter;
874 if (backer != ofproto->backer) {
878 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
881 if (!iter->tnl_port) {
885 dp_port = netdev_vport_get_dpif_port(iter->up.netdev);
886 node = simap_find(&tmp_backers, dp_port);
888 simap_put(&backer->tnl_backers, dp_port, node->data);
889 simap_delete(&tmp_backers, node);
890 node = simap_find(&backer->tnl_backers, dp_port);
892 node = simap_find(&backer->tnl_backers, dp_port);
894 uint32_t odp_port = UINT32_MAX;
896 if (!dpif_port_add(backer->dpif, iter->up.netdev,
898 simap_put(&backer->tnl_backers, dp_port, odp_port);
899 node = simap_find(&backer->tnl_backers, dp_port);
904 iter->odp_port = node ? node->data : OVSP_NONE;
905 if (tnl_port_reconfigure(&iter->up, iter->odp_port,
907 backer->need_revalidate = REV_RECONFIGURE;
912 SIMAP_FOR_EACH (node, &tmp_backers) {
913 dpif_port_del(backer->dpif, node->data);
915 simap_destroy(&tmp_backers);
917 switch (backer->need_revalidate) {
918 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
919 case REV_STP: COVERAGE_INC(rev_stp); break;
920 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
921 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
922 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
925 if (backer->need_revalidate) {
926 /* Clear the drop_keys in case we should now be accepting some
927 * formerly dropped flows. */
928 drop_key_clear(backer);
931 /* Clear the revalidation flags. */
932 tag_set_init(&backer->revalidate_set);
933 backer->need_revalidate = 0;
935 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
936 struct facet *facet, *next;
938 if (ofproto->backer != backer) {
942 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
944 || tag_set_intersects(&revalidate_set, facet->tags)) {
945 facet_revalidate(facet);
951 if (timer_expired(&backer->next_expiration)) {
952 int delay = expire(backer);
953 timer_set_duration(&backer->next_expiration, delay);
956 /* Check for port changes in the dpif. */
957 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
958 struct ofproto_dpif *ofproto;
959 struct dpif_port port;
961 /* Don't report on the datapath's device. */
962 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
966 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
967 &all_ofproto_dpifs) {
968 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
973 ofproto = lookup_ofproto_dpif_by_port_name(devname);
974 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
975 /* The port was removed. If we know the datapath,
976 * report it through poll_set(). If we don't, it may be
977 * notifying us of a removal we initiated, so ignore it.
978 * If there's a pending ENOBUFS, let it stand, since
979 * everything will be reevaluated. */
980 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
981 sset_add(&ofproto->port_poll_set, devname);
982 ofproto->port_poll_errno = 0;
984 } else if (!ofproto) {
985 /* The port was added, but we don't know with which
986 * ofproto we should associate it. Delete it. */
987 dpif_port_del(backer->dpif, port.port_no);
989 dpif_port_destroy(&port);
995 if (error != EAGAIN) {
996 struct ofproto_dpif *ofproto;
998 /* There was some sort of error, so propagate it to all
999 * ofprotos that use this backer. */
1000 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1001 &all_ofproto_dpifs) {
1002 if (ofproto->backer == backer) {
1003 sset_clear(&ofproto->port_poll_set);
1004 ofproto->port_poll_errno = error;
1013 type_run_fast(const char *type)
1015 struct dpif_backer *backer;
1018 backer = shash_find_data(&all_dpif_backers, type);
1020 /* This is not necessarily a problem, since backers are only
1021 * created on demand. */
1025 /* Handle one or more batches of upcalls, until there's nothing left to do
1026 * or until we do a fixed total amount of work.
1028 * We do work in batches because it can be much cheaper to set up a number
1029 * of flows and fire off their patches all at once. We do multiple batches
1030 * because in some cases handling a packet can cause another packet to be
1031 * queued almost immediately as part of the return flow. Both
1032 * optimizations can make major improvements on some benchmarks and
1033 * presumably for real traffic as well. */
1035 while (work < FLOW_MISS_MAX_BATCH) {
1036 int retval = handle_upcalls(backer, FLOW_MISS_MAX_BATCH - work);
1047 type_wait(const char *type)
1049 struct dpif_backer *backer;
1051 backer = shash_find_data(&all_dpif_backers, type);
1053 /* This is not necessarily a problem, since backers are only
1054 * created on demand. */
1058 timer_wait(&backer->next_expiration);
1061 /* Basic life-cycle. */
1063 static int add_internal_flows(struct ofproto_dpif *);
1065 static struct ofproto *
1068 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1069 return &ofproto->up;
1073 dealloc(struct ofproto *ofproto_)
1075 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1080 close_dpif_backer(struct dpif_backer *backer)
1082 struct shash_node *node;
1084 ovs_assert(backer->refcount > 0);
1086 if (--backer->refcount) {
1090 drop_key_clear(backer);
1091 hmap_destroy(&backer->drop_keys);
1093 simap_destroy(&backer->tnl_backers);
1094 hmap_destroy(&backer->odp_to_ofport_map);
1095 node = shash_find(&all_dpif_backers, backer->type);
1097 shash_delete(&all_dpif_backers, node);
1098 dpif_close(backer->dpif);
1103 /* Datapath port slated for removal from datapath. */
1104 struct odp_garbage {
1105 struct list list_node;
1110 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1112 struct dpif_backer *backer;
1113 struct dpif_port_dump port_dump;
1114 struct dpif_port port;
1115 struct shash_node *node;
1116 struct list garbage_list;
1117 struct odp_garbage *garbage, *next;
1123 backer = shash_find_data(&all_dpif_backers, type);
1130 backer_name = xasprintf("ovs-%s", type);
1132 /* Remove any existing datapaths, since we assume we're the only
1133 * userspace controlling the datapath. */
1135 dp_enumerate_names(type, &names);
1136 SSET_FOR_EACH(name, &names) {
1137 struct dpif *old_dpif;
1139 /* Don't remove our backer if it exists. */
1140 if (!strcmp(name, backer_name)) {
1144 if (dpif_open(name, type, &old_dpif)) {
1145 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1147 dpif_delete(old_dpif);
1148 dpif_close(old_dpif);
1151 sset_destroy(&names);
1153 backer = xmalloc(sizeof *backer);
1155 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1158 VLOG_ERR("failed to open datapath of type %s: %s", type,
1164 backer->type = xstrdup(type);
1165 backer->refcount = 1;
1166 hmap_init(&backer->odp_to_ofport_map);
1167 hmap_init(&backer->drop_keys);
1168 timer_set_duration(&backer->next_expiration, 1000);
1169 backer->need_revalidate = 0;
1170 simap_init(&backer->tnl_backers);
1171 tag_set_init(&backer->revalidate_set);
1174 dpif_flow_flush(backer->dpif);
1176 /* Loop through the ports already on the datapath and remove any
1177 * that we don't need anymore. */
1178 list_init(&garbage_list);
1179 dpif_port_dump_start(&port_dump, backer->dpif);
1180 while (dpif_port_dump_next(&port_dump, &port)) {
1181 node = shash_find(&init_ofp_ports, port.name);
1182 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1183 garbage = xmalloc(sizeof *garbage);
1184 garbage->odp_port = port.port_no;
1185 list_push_front(&garbage_list, &garbage->list_node);
1188 dpif_port_dump_done(&port_dump);
1190 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1191 dpif_port_del(backer->dpif, garbage->odp_port);
1192 list_remove(&garbage->list_node);
1196 shash_add(&all_dpif_backers, type, backer);
1198 error = dpif_recv_set(backer->dpif, true);
1200 VLOG_ERR("failed to listen on datapath of type %s: %s",
1201 type, strerror(error));
1202 close_dpif_backer(backer);
1210 construct(struct ofproto *ofproto_)
1212 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1213 struct shash_node *node, *next;
1218 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1223 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1224 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1226 ofproto->n_matches = 0;
1228 ofproto->netflow = NULL;
1229 ofproto->sflow = NULL;
1230 ofproto->stp = NULL;
1231 hmap_init(&ofproto->bundles);
1232 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1233 for (i = 0; i < MAX_MIRRORS; i++) {
1234 ofproto->mirrors[i] = NULL;
1236 ofproto->has_bonded_bundles = false;
1238 hmap_init(&ofproto->facets);
1239 hmap_init(&ofproto->subfacets);
1240 ofproto->governor = NULL;
1241 ofproto->consistency_rl = LLONG_MIN;
1243 for (i = 0; i < N_TABLES; i++) {
1244 struct table_dpif *table = &ofproto->tables[i];
1246 table->catchall_table = NULL;
1247 table->other_table = NULL;
1248 table->basis = random_uint32();
1251 list_init(&ofproto->completions);
1253 ofproto_dpif_unixctl_init();
1255 ofproto->has_mirrors = false;
1256 ofproto->has_bundle_action = false;
1258 hmap_init(&ofproto->vlandev_map);
1259 hmap_init(&ofproto->realdev_vid_map);
1261 sset_init(&ofproto->ports);
1262 sset_init(&ofproto->ghost_ports);
1263 sset_init(&ofproto->port_poll_set);
1264 ofproto->port_poll_errno = 0;
1266 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1267 struct iface_hint *iface_hint = node->data;
1269 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1270 /* Check if the datapath already has this port. */
1271 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1272 sset_add(&ofproto->ports, node->name);
1275 free(iface_hint->br_name);
1276 free(iface_hint->br_type);
1278 shash_delete(&init_ofp_ports, node);
1282 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1283 hash_string(ofproto->up.name, 0));
1284 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1286 ofproto_init_tables(ofproto_, N_TABLES);
1287 error = add_internal_flows(ofproto);
1288 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1294 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1295 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1297 struct ofputil_flow_mod fm;
1300 match_init_catchall(&fm.match);
1302 match_set_reg(&fm.match, 0, id);
1303 fm.new_cookie = htonll(0);
1304 fm.cookie = htonll(0);
1305 fm.cookie_mask = htonll(0);
1306 fm.table_id = TBL_INTERNAL;
1307 fm.command = OFPFC_ADD;
1308 fm.idle_timeout = 0;
1309 fm.hard_timeout = 0;
1313 fm.ofpacts = ofpacts->data;
1314 fm.ofpacts_len = ofpacts->size;
1316 error = ofproto_flow_mod(&ofproto->up, &fm);
1318 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1319 id, ofperr_to_string(error));
1323 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1324 ovs_assert(*rulep != NULL);
1330 add_internal_flows(struct ofproto_dpif *ofproto)
1332 struct ofpact_controller *controller;
1333 uint64_t ofpacts_stub[128 / 8];
1334 struct ofpbuf ofpacts;
1338 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1341 controller = ofpact_put_CONTROLLER(&ofpacts);
1342 controller->max_len = UINT16_MAX;
1343 controller->controller_id = 0;
1344 controller->reason = OFPR_NO_MATCH;
1345 ofpact_pad(&ofpacts);
1347 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1352 ofpbuf_clear(&ofpacts);
1353 error = add_internal_flow(ofproto, id++, &ofpacts,
1354 &ofproto->no_packet_in_rule);
1359 complete_operations(struct ofproto_dpif *ofproto)
1361 struct dpif_completion *c, *next;
1363 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1364 ofoperation_complete(c->op, 0);
1365 list_remove(&c->list_node);
1371 destruct(struct ofproto *ofproto_)
1373 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1374 struct rule_dpif *rule, *next_rule;
1375 struct oftable *table;
1378 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1379 complete_operations(ofproto);
1381 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1382 struct cls_cursor cursor;
1384 cls_cursor_init(&cursor, &table->cls, NULL);
1385 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1386 ofproto_rule_destroy(&rule->up);
1390 for (i = 0; i < MAX_MIRRORS; i++) {
1391 mirror_destroy(ofproto->mirrors[i]);
1394 netflow_destroy(ofproto->netflow);
1395 dpif_sflow_destroy(ofproto->sflow);
1396 hmap_destroy(&ofproto->bundles);
1397 mac_learning_destroy(ofproto->ml);
1399 hmap_destroy(&ofproto->facets);
1400 hmap_destroy(&ofproto->subfacets);
1401 governor_destroy(ofproto->governor);
1403 hmap_destroy(&ofproto->vlandev_map);
1404 hmap_destroy(&ofproto->realdev_vid_map);
1406 sset_destroy(&ofproto->ports);
1407 sset_destroy(&ofproto->ghost_ports);
1408 sset_destroy(&ofproto->port_poll_set);
1410 close_dpif_backer(ofproto->backer);
1414 run_fast(struct ofproto *ofproto_)
1416 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1417 struct ofport_dpif *ofport;
1419 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1420 port_run_fast(ofport);
1427 run(struct ofproto *ofproto_)
1429 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1430 struct ofport_dpif *ofport;
1431 struct ofbundle *bundle;
1435 complete_operations(ofproto);
1438 error = run_fast(ofproto_);
1443 if (ofproto->netflow) {
1444 if (netflow_run(ofproto->netflow)) {
1445 send_netflow_active_timeouts(ofproto);
1448 if (ofproto->sflow) {
1449 dpif_sflow_run(ofproto->sflow);
1452 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1455 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1460 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1462 /* Check the consistency of a random facet, to aid debugging. */
1463 if (time_msec() >= ofproto->consistency_rl
1464 && !hmap_is_empty(&ofproto->facets)
1465 && !ofproto->backer->need_revalidate) {
1466 struct facet *facet;
1468 ofproto->consistency_rl = time_msec() + 250;
1470 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1471 struct facet, hmap_node);
1472 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1474 if (!facet_check_consistency(facet)) {
1475 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1480 if (ofproto->governor) {
1483 governor_run(ofproto->governor);
1485 /* If the governor has shrunk to its minimum size and the number of
1486 * subfacets has dwindled, then drop the governor entirely.
1488 * For hysteresis, the number of subfacets to drop the governor is
1489 * smaller than the number needed to trigger its creation. */
1490 n_subfacets = hmap_count(&ofproto->subfacets);
1491 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1492 && governor_is_idle(ofproto->governor)) {
1493 governor_destroy(ofproto->governor);
1494 ofproto->governor = NULL;
1502 wait(struct ofproto *ofproto_)
1504 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1505 struct ofport_dpif *ofport;
1506 struct ofbundle *bundle;
1508 if (!clogged && !list_is_empty(&ofproto->completions)) {
1509 poll_immediate_wake();
1512 dpif_wait(ofproto->backer->dpif);
1513 dpif_recv_wait(ofproto->backer->dpif);
1514 if (ofproto->sflow) {
1515 dpif_sflow_wait(ofproto->sflow);
1517 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1518 poll_immediate_wake();
1520 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1523 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1524 bundle_wait(bundle);
1526 if (ofproto->netflow) {
1527 netflow_wait(ofproto->netflow);
1529 mac_learning_wait(ofproto->ml);
1531 if (ofproto->backer->need_revalidate) {
1532 /* Shouldn't happen, but if it does just go around again. */
1533 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1534 poll_immediate_wake();
1536 if (ofproto->governor) {
1537 governor_wait(ofproto->governor);
1542 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1544 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1546 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1547 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1551 flush(struct ofproto *ofproto_)
1553 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1554 struct subfacet *subfacet, *next_subfacet;
1555 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1559 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1560 &ofproto->subfacets) {
1561 if (subfacet->path != SF_NOT_INSTALLED) {
1562 batch[n_batch++] = subfacet;
1563 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1564 subfacet_destroy_batch(ofproto, batch, n_batch);
1568 subfacet_destroy(subfacet);
1573 subfacet_destroy_batch(ofproto, batch, n_batch);
1578 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1579 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1581 *arp_match_ip = true;
1582 *actions = (OFPUTIL_A_OUTPUT |
1583 OFPUTIL_A_SET_VLAN_VID |
1584 OFPUTIL_A_SET_VLAN_PCP |
1585 OFPUTIL_A_STRIP_VLAN |
1586 OFPUTIL_A_SET_DL_SRC |
1587 OFPUTIL_A_SET_DL_DST |
1588 OFPUTIL_A_SET_NW_SRC |
1589 OFPUTIL_A_SET_NW_DST |
1590 OFPUTIL_A_SET_NW_TOS |
1591 OFPUTIL_A_SET_TP_SRC |
1592 OFPUTIL_A_SET_TP_DST |
1597 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1599 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1600 struct dpif_dp_stats s;
1602 strcpy(ots->name, "classifier");
1604 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1606 ots->lookup_count = htonll(s.n_hit + s.n_missed);
1607 ots->matched_count = htonll(s.n_hit + ofproto->n_matches);
1610 static struct ofport *
1613 struct ofport_dpif *port = xmalloc(sizeof *port);
1618 port_dealloc(struct ofport *port_)
1620 struct ofport_dpif *port = ofport_dpif_cast(port_);
1625 port_construct(struct ofport *port_)
1627 struct ofport_dpif *port = ofport_dpif_cast(port_);
1628 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1629 const struct netdev *netdev = port->up.netdev;
1630 struct dpif_port dpif_port;
1633 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1634 port->bundle = NULL;
1636 port->tag = tag_create_random();
1637 port->may_enable = true;
1638 port->stp_port = NULL;
1639 port->stp_state = STP_DISABLED;
1640 port->tnl_port = NULL;
1641 hmap_init(&port->priorities);
1642 port->realdev_ofp_port = 0;
1643 port->vlandev_vid = 0;
1644 port->carrier_seq = netdev_get_carrier_resets(netdev);
1646 if (netdev_vport_is_patch(netdev)) {
1647 /* XXX By bailing out here, we don't do required sFlow work. */
1648 port->odp_port = OVSP_NONE;
1652 error = dpif_port_query_by_name(ofproto->backer->dpif,
1653 netdev_vport_get_dpif_port(netdev),
1659 port->odp_port = dpif_port.port_no;
1661 if (netdev_get_tunnel_config(netdev)) {
1662 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1664 /* Sanity-check that a mapping doesn't already exist. This
1665 * shouldn't happen for non-tunnel ports. */
1666 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1667 VLOG_ERR("port %s already has an OpenFlow port number",
1669 dpif_port_destroy(&dpif_port);
1673 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1674 hash_int(port->odp_port, 0));
1676 dpif_port_destroy(&dpif_port);
1678 if (ofproto->sflow) {
1679 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1686 port_destruct(struct ofport *port_)
1688 struct ofport_dpif *port = ofport_dpif_cast(port_);
1689 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1690 const char *dp_port_name = netdev_vport_get_dpif_port(port->up.netdev);
1691 const char *devname = netdev_get_name(port->up.netdev);
1693 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1694 /* The underlying device is still there, so delete it. This
1695 * happens when the ofproto is being destroyed, since the caller
1696 * assumes that removal of attached ports will happen as part of
1698 if (!port->tnl_port) {
1699 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1701 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1704 if (port->odp_port != OVSP_NONE && !port->tnl_port) {
1705 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1708 tnl_port_del(port->tnl_port);
1709 sset_find_and_delete(&ofproto->ports, devname);
1710 sset_find_and_delete(&ofproto->ghost_ports, devname);
1711 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1712 bundle_remove(port_);
1713 set_cfm(port_, NULL);
1714 if (ofproto->sflow) {
1715 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1718 ofport_clear_priorities(port);
1719 hmap_destroy(&port->priorities);
1723 port_modified(struct ofport *port_)
1725 struct ofport_dpif *port = ofport_dpif_cast(port_);
1727 if (port->bundle && port->bundle->bond) {
1728 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1733 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1735 struct ofport_dpif *port = ofport_dpif_cast(port_);
1736 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1737 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1739 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1740 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1741 OFPUTIL_PC_NO_PACKET_IN)) {
1742 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1744 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1745 bundle_update(port->bundle);
1751 set_sflow(struct ofproto *ofproto_,
1752 const struct ofproto_sflow_options *sflow_options)
1754 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1755 struct dpif_sflow *ds = ofproto->sflow;
1757 if (sflow_options) {
1759 struct ofport_dpif *ofport;
1761 ds = ofproto->sflow = dpif_sflow_create();
1762 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1763 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1765 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1767 dpif_sflow_set_options(ds, sflow_options);
1770 dpif_sflow_destroy(ds);
1771 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1772 ofproto->sflow = NULL;
1779 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1781 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1788 struct ofproto_dpif *ofproto;
1790 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1791 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1792 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1795 if (cfm_configure(ofport->cfm, s)) {
1801 cfm_destroy(ofport->cfm);
1807 get_cfm_fault(const struct ofport *ofport_)
1809 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1811 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1815 get_cfm_opup(const struct ofport *ofport_)
1817 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1819 return ofport->cfm ? cfm_get_opup(ofport->cfm) : -1;
1823 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1826 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1829 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1837 get_cfm_health(const struct ofport *ofport_)
1839 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1841 return ofport->cfm ? cfm_get_health(ofport->cfm) : -1;
1844 /* Spanning Tree. */
1847 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1849 struct ofproto_dpif *ofproto = ofproto_;
1850 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1851 struct ofport_dpif *ofport;
1853 ofport = stp_port_get_aux(sp);
1855 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1856 ofproto->up.name, port_num);
1858 struct eth_header *eth = pkt->l2;
1860 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1861 if (eth_addr_is_zero(eth->eth_src)) {
1862 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1863 "with unknown MAC", ofproto->up.name, port_num);
1865 send_packet(ofport, pkt);
1871 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1873 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1875 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1877 /* Only revalidate flows if the configuration changed. */
1878 if (!s != !ofproto->stp) {
1879 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1883 if (!ofproto->stp) {
1884 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1885 send_bpdu_cb, ofproto);
1886 ofproto->stp_last_tick = time_msec();
1889 stp_set_bridge_id(ofproto->stp, s->system_id);
1890 stp_set_bridge_priority(ofproto->stp, s->priority);
1891 stp_set_hello_time(ofproto->stp, s->hello_time);
1892 stp_set_max_age(ofproto->stp, s->max_age);
1893 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1895 struct ofport *ofport;
1897 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1898 set_stp_port(ofport, NULL);
1901 stp_destroy(ofproto->stp);
1902 ofproto->stp = NULL;
1909 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1911 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1915 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1916 s->designated_root = stp_get_designated_root(ofproto->stp);
1917 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1926 update_stp_port_state(struct ofport_dpif *ofport)
1928 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1929 enum stp_state state;
1931 /* Figure out new state. */
1932 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1936 if (ofport->stp_state != state) {
1937 enum ofputil_port_state of_state;
1940 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1941 netdev_get_name(ofport->up.netdev),
1942 stp_state_name(ofport->stp_state),
1943 stp_state_name(state));
1944 if (stp_learn_in_state(ofport->stp_state)
1945 != stp_learn_in_state(state)) {
1946 /* xxx Learning action flows should also be flushed. */
1947 mac_learning_flush(ofproto->ml,
1948 &ofproto->backer->revalidate_set);
1950 fwd_change = stp_forward_in_state(ofport->stp_state)
1951 != stp_forward_in_state(state);
1953 ofproto->backer->need_revalidate = REV_STP;
1954 ofport->stp_state = state;
1955 ofport->stp_state_entered = time_msec();
1957 if (fwd_change && ofport->bundle) {
1958 bundle_update(ofport->bundle);
1961 /* Update the STP state bits in the OpenFlow port description. */
1962 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1963 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1964 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1965 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1966 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1968 ofproto_port_set_state(&ofport->up, of_state);
1972 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1973 * caller is responsible for assigning STP port numbers and ensuring
1974 * there are no duplicates. */
1976 set_stp_port(struct ofport *ofport_,
1977 const struct ofproto_port_stp_settings *s)
1979 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1980 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1981 struct stp_port *sp = ofport->stp_port;
1983 if (!s || !s->enable) {
1985 ofport->stp_port = NULL;
1986 stp_port_disable(sp);
1987 update_stp_port_state(ofport);
1990 } else if (sp && stp_port_no(sp) != s->port_num
1991 && ofport == stp_port_get_aux(sp)) {
1992 /* The port-id changed, so disable the old one if it's not
1993 * already in use by another port. */
1994 stp_port_disable(sp);
1997 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1998 stp_port_enable(sp);
2000 stp_port_set_aux(sp, ofport);
2001 stp_port_set_priority(sp, s->priority);
2002 stp_port_set_path_cost(sp, s->path_cost);
2004 update_stp_port_state(ofport);
2010 get_stp_port_status(struct ofport *ofport_,
2011 struct ofproto_port_stp_status *s)
2013 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2014 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2015 struct stp_port *sp = ofport->stp_port;
2017 if (!ofproto->stp || !sp) {
2023 s->port_id = stp_port_get_id(sp);
2024 s->state = stp_port_get_state(sp);
2025 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2026 s->role = stp_port_get_role(sp);
2027 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2033 stp_run(struct ofproto_dpif *ofproto)
2036 long long int now = time_msec();
2037 long long int elapsed = now - ofproto->stp_last_tick;
2038 struct stp_port *sp;
2041 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2042 ofproto->stp_last_tick = now;
2044 while (stp_get_changed_port(ofproto->stp, &sp)) {
2045 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2048 update_stp_port_state(ofport);
2052 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2053 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2059 stp_wait(struct ofproto_dpif *ofproto)
2062 poll_timer_wait(1000);
2066 /* Returns true if STP should process 'flow'. */
2068 stp_should_process_flow(const struct flow *flow)
2070 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2074 stp_process_packet(const struct ofport_dpif *ofport,
2075 const struct ofpbuf *packet)
2077 struct ofpbuf payload = *packet;
2078 struct eth_header *eth = payload.data;
2079 struct stp_port *sp = ofport->stp_port;
2081 /* Sink packets on ports that have STP disabled when the bridge has
2083 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2087 /* Trim off padding on payload. */
2088 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2089 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2092 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2093 stp_received_bpdu(sp, payload.data, payload.size);
2097 static struct priority_to_dscp *
2098 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2100 struct priority_to_dscp *pdscp;
2103 hash = hash_int(priority, 0);
2104 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2105 if (pdscp->priority == priority) {
2113 ofport_clear_priorities(struct ofport_dpif *ofport)
2115 struct priority_to_dscp *pdscp, *next;
2117 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2118 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2124 set_queues(struct ofport *ofport_,
2125 const struct ofproto_port_queue *qdscp_list,
2128 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2129 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2130 struct hmap new = HMAP_INITIALIZER(&new);
2133 for (i = 0; i < n_qdscp; i++) {
2134 struct priority_to_dscp *pdscp;
2138 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2139 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2144 pdscp = get_priority(ofport, priority);
2146 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2148 pdscp = xmalloc(sizeof *pdscp);
2149 pdscp->priority = priority;
2151 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2154 if (pdscp->dscp != dscp) {
2156 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2159 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2162 if (!hmap_is_empty(&ofport->priorities)) {
2163 ofport_clear_priorities(ofport);
2164 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2167 hmap_swap(&new, &ofport->priorities);
2175 /* Expires all MAC learning entries associated with 'bundle' and forces its
2176 * ofproto to revalidate every flow.
2178 * Normally MAC learning entries are removed only from the ofproto associated
2179 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2180 * are removed from every ofproto. When patch ports and SLB bonds are in use
2181 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2182 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2183 * with the host from which it migrated. */
2185 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2187 struct ofproto_dpif *ofproto = bundle->ofproto;
2188 struct mac_learning *ml = ofproto->ml;
2189 struct mac_entry *mac, *next_mac;
2191 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2192 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2193 if (mac->port.p == bundle) {
2195 struct ofproto_dpif *o;
2197 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2199 struct mac_entry *e;
2201 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2204 mac_learning_expire(o->ml, e);
2210 mac_learning_expire(ml, mac);
2215 static struct ofbundle *
2216 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2218 struct ofbundle *bundle;
2220 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2221 &ofproto->bundles) {
2222 if (bundle->aux == aux) {
2229 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2230 * ones that are found to 'bundles'. */
2232 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2233 void **auxes, size_t n_auxes,
2234 struct hmapx *bundles)
2238 hmapx_init(bundles);
2239 for (i = 0; i < n_auxes; i++) {
2240 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2242 hmapx_add(bundles, bundle);
2248 bundle_update(struct ofbundle *bundle)
2250 struct ofport_dpif *port;
2252 bundle->floodable = true;
2253 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2254 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2255 || !stp_forward_in_state(port->stp_state)) {
2256 bundle->floodable = false;
2263 bundle_del_port(struct ofport_dpif *port)
2265 struct ofbundle *bundle = port->bundle;
2267 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2269 list_remove(&port->bundle_node);
2270 port->bundle = NULL;
2273 lacp_slave_unregister(bundle->lacp, port);
2276 bond_slave_unregister(bundle->bond, port);
2279 bundle_update(bundle);
2283 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
2284 struct lacp_slave_settings *lacp,
2285 uint32_t bond_stable_id)
2287 struct ofport_dpif *port;
2289 port = get_ofp_port(bundle->ofproto, ofp_port);
2294 if (port->bundle != bundle) {
2295 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2297 bundle_del_port(port);
2300 port->bundle = bundle;
2301 list_push_back(&bundle->ports, &port->bundle_node);
2302 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2303 || !stp_forward_in_state(port->stp_state)) {
2304 bundle->floodable = false;
2308 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2309 lacp_slave_register(bundle->lacp, port, lacp);
2312 port->bond_stable_id = bond_stable_id;
2318 bundle_destroy(struct ofbundle *bundle)
2320 struct ofproto_dpif *ofproto;
2321 struct ofport_dpif *port, *next_port;
2328 ofproto = bundle->ofproto;
2329 for (i = 0; i < MAX_MIRRORS; i++) {
2330 struct ofmirror *m = ofproto->mirrors[i];
2332 if (m->out == bundle) {
2334 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2335 || hmapx_find_and_delete(&m->dsts, bundle)) {
2336 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2341 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2342 bundle_del_port(port);
2345 bundle_flush_macs(bundle, true);
2346 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2348 free(bundle->trunks);
2349 lacp_destroy(bundle->lacp);
2350 bond_destroy(bundle->bond);
2355 bundle_set(struct ofproto *ofproto_, void *aux,
2356 const struct ofproto_bundle_settings *s)
2358 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2359 bool need_flush = false;
2360 struct ofport_dpif *port;
2361 struct ofbundle *bundle;
2362 unsigned long *trunks;
2368 bundle_destroy(bundle_lookup(ofproto, aux));
2372 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2373 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2375 bundle = bundle_lookup(ofproto, aux);
2377 bundle = xmalloc(sizeof *bundle);
2379 bundle->ofproto = ofproto;
2380 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2381 hash_pointer(aux, 0));
2383 bundle->name = NULL;
2385 list_init(&bundle->ports);
2386 bundle->vlan_mode = PORT_VLAN_TRUNK;
2388 bundle->trunks = NULL;
2389 bundle->use_priority_tags = s->use_priority_tags;
2390 bundle->lacp = NULL;
2391 bundle->bond = NULL;
2393 bundle->floodable = true;
2395 bundle->src_mirrors = 0;
2396 bundle->dst_mirrors = 0;
2397 bundle->mirror_out = 0;
2400 if (!bundle->name || strcmp(s->name, bundle->name)) {
2402 bundle->name = xstrdup(s->name);
2407 if (!bundle->lacp) {
2408 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2409 bundle->lacp = lacp_create();
2411 lacp_configure(bundle->lacp, s->lacp);
2413 lacp_destroy(bundle->lacp);
2414 bundle->lacp = NULL;
2417 /* Update set of ports. */
2419 for (i = 0; i < s->n_slaves; i++) {
2420 if (!bundle_add_port(bundle, s->slaves[i],
2421 s->lacp ? &s->lacp_slaves[i] : NULL,
2422 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
2426 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2427 struct ofport_dpif *next_port;
2429 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2430 for (i = 0; i < s->n_slaves; i++) {
2431 if (s->slaves[i] == port->up.ofp_port) {
2436 bundle_del_port(port);
2440 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2442 if (list_is_empty(&bundle->ports)) {
2443 bundle_destroy(bundle);
2447 /* Set VLAN tagging mode */
2448 if (s->vlan_mode != bundle->vlan_mode
2449 || s->use_priority_tags != bundle->use_priority_tags) {
2450 bundle->vlan_mode = s->vlan_mode;
2451 bundle->use_priority_tags = s->use_priority_tags;
2456 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2457 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2459 if (vlan != bundle->vlan) {
2460 bundle->vlan = vlan;
2464 /* Get trunked VLANs. */
2465 switch (s->vlan_mode) {
2466 case PORT_VLAN_ACCESS:
2470 case PORT_VLAN_TRUNK:
2471 trunks = CONST_CAST(unsigned long *, s->trunks);
2474 case PORT_VLAN_NATIVE_UNTAGGED:
2475 case PORT_VLAN_NATIVE_TAGGED:
2476 if (vlan != 0 && (!s->trunks
2477 || !bitmap_is_set(s->trunks, vlan)
2478 || bitmap_is_set(s->trunks, 0))) {
2479 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2481 trunks = bitmap_clone(s->trunks, 4096);
2483 trunks = bitmap_allocate1(4096);
2485 bitmap_set1(trunks, vlan);
2486 bitmap_set0(trunks, 0);
2488 trunks = CONST_CAST(unsigned long *, s->trunks);
2495 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2496 free(bundle->trunks);
2497 if (trunks == s->trunks) {
2498 bundle->trunks = vlan_bitmap_clone(trunks);
2500 bundle->trunks = trunks;
2505 if (trunks != s->trunks) {
2510 if (!list_is_short(&bundle->ports)) {
2511 bundle->ofproto->has_bonded_bundles = true;
2513 if (bond_reconfigure(bundle->bond, s->bond)) {
2514 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2517 bundle->bond = bond_create(s->bond);
2518 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2521 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2522 bond_slave_register(bundle->bond, port, port->bond_stable_id,
2526 bond_destroy(bundle->bond);
2527 bundle->bond = NULL;
2530 /* If we changed something that would affect MAC learning, un-learn
2531 * everything on this port and force flow revalidation. */
2533 bundle_flush_macs(bundle, false);
2540 bundle_remove(struct ofport *port_)
2542 struct ofport_dpif *port = ofport_dpif_cast(port_);
2543 struct ofbundle *bundle = port->bundle;
2546 bundle_del_port(port);
2547 if (list_is_empty(&bundle->ports)) {
2548 bundle_destroy(bundle);
2549 } else if (list_is_short(&bundle->ports)) {
2550 bond_destroy(bundle->bond);
2551 bundle->bond = NULL;
2557 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2559 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2560 struct ofport_dpif *port = port_;
2561 uint8_t ea[ETH_ADDR_LEN];
2564 error = netdev_get_etheraddr(port->up.netdev, ea);
2566 struct ofpbuf packet;
2569 ofpbuf_init(&packet, 0);
2570 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2572 memcpy(packet_pdu, pdu, pdu_size);
2574 send_packet(port, &packet);
2575 ofpbuf_uninit(&packet);
2577 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2578 "%s (%s)", port->bundle->name,
2579 netdev_get_name(port->up.netdev), strerror(error));
2584 bundle_send_learning_packets(struct ofbundle *bundle)
2586 struct ofproto_dpif *ofproto = bundle->ofproto;
2587 int error, n_packets, n_errors;
2588 struct mac_entry *e;
2590 error = n_packets = n_errors = 0;
2591 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2592 if (e->port.p != bundle) {
2593 struct ofpbuf *learning_packet;
2594 struct ofport_dpif *port;
2598 /* The assignment to "port" is unnecessary but makes "grep"ing for
2599 * struct ofport_dpif more effective. */
2600 learning_packet = bond_compose_learning_packet(bundle->bond,
2604 ret = send_packet(port, learning_packet);
2605 ofpbuf_delete(learning_packet);
2615 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2616 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2617 "packets, last error was: %s",
2618 bundle->name, n_errors, n_packets, strerror(error));
2620 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2621 bundle->name, n_packets);
2626 bundle_run(struct ofbundle *bundle)
2629 lacp_run(bundle->lacp, send_pdu_cb);
2632 struct ofport_dpif *port;
2634 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2635 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2638 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2639 lacp_status(bundle->lacp));
2640 if (bond_should_send_learning_packets(bundle->bond)) {
2641 bundle_send_learning_packets(bundle);
2647 bundle_wait(struct ofbundle *bundle)
2650 lacp_wait(bundle->lacp);
2653 bond_wait(bundle->bond);
2660 mirror_scan(struct ofproto_dpif *ofproto)
2664 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2665 if (!ofproto->mirrors[idx]) {
2672 static struct ofmirror *
2673 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2677 for (i = 0; i < MAX_MIRRORS; i++) {
2678 struct ofmirror *mirror = ofproto->mirrors[i];
2679 if (mirror && mirror->aux == aux) {
2687 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2689 mirror_update_dups(struct ofproto_dpif *ofproto)
2693 for (i = 0; i < MAX_MIRRORS; i++) {
2694 struct ofmirror *m = ofproto->mirrors[i];
2697 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2701 for (i = 0; i < MAX_MIRRORS; i++) {
2702 struct ofmirror *m1 = ofproto->mirrors[i];
2709 for (j = i + 1; j < MAX_MIRRORS; j++) {
2710 struct ofmirror *m2 = ofproto->mirrors[j];
2712 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2713 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2714 m2->dup_mirrors |= m1->dup_mirrors;
2721 mirror_set(struct ofproto *ofproto_, void *aux,
2722 const struct ofproto_mirror_settings *s)
2724 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2725 mirror_mask_t mirror_bit;
2726 struct ofbundle *bundle;
2727 struct ofmirror *mirror;
2728 struct ofbundle *out;
2729 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2730 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2733 mirror = mirror_lookup(ofproto, aux);
2735 mirror_destroy(mirror);
2741 idx = mirror_scan(ofproto);
2743 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2745 ofproto->up.name, MAX_MIRRORS, s->name);
2749 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2750 mirror->ofproto = ofproto;
2753 mirror->out_vlan = -1;
2754 mirror->name = NULL;
2757 if (!mirror->name || strcmp(s->name, mirror->name)) {
2759 mirror->name = xstrdup(s->name);
2762 /* Get the new configuration. */
2763 if (s->out_bundle) {
2764 out = bundle_lookup(ofproto, s->out_bundle);
2766 mirror_destroy(mirror);
2772 out_vlan = s->out_vlan;
2774 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2775 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2777 /* If the configuration has not changed, do nothing. */
2778 if (hmapx_equals(&srcs, &mirror->srcs)
2779 && hmapx_equals(&dsts, &mirror->dsts)
2780 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2781 && mirror->out == out
2782 && mirror->out_vlan == out_vlan)
2784 hmapx_destroy(&srcs);
2785 hmapx_destroy(&dsts);
2789 hmapx_swap(&srcs, &mirror->srcs);
2790 hmapx_destroy(&srcs);
2792 hmapx_swap(&dsts, &mirror->dsts);
2793 hmapx_destroy(&dsts);
2795 free(mirror->vlans);
2796 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2799 mirror->out_vlan = out_vlan;
2801 /* Update bundles. */
2802 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2803 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2804 if (hmapx_contains(&mirror->srcs, bundle)) {
2805 bundle->src_mirrors |= mirror_bit;
2807 bundle->src_mirrors &= ~mirror_bit;
2810 if (hmapx_contains(&mirror->dsts, bundle)) {
2811 bundle->dst_mirrors |= mirror_bit;
2813 bundle->dst_mirrors &= ~mirror_bit;
2816 if (mirror->out == bundle) {
2817 bundle->mirror_out |= mirror_bit;
2819 bundle->mirror_out &= ~mirror_bit;
2823 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2824 ofproto->has_mirrors = true;
2825 mac_learning_flush(ofproto->ml,
2826 &ofproto->backer->revalidate_set);
2827 mirror_update_dups(ofproto);
2833 mirror_destroy(struct ofmirror *mirror)
2835 struct ofproto_dpif *ofproto;
2836 mirror_mask_t mirror_bit;
2837 struct ofbundle *bundle;
2844 ofproto = mirror->ofproto;
2845 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2846 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2848 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2849 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2850 bundle->src_mirrors &= ~mirror_bit;
2851 bundle->dst_mirrors &= ~mirror_bit;
2852 bundle->mirror_out &= ~mirror_bit;
2855 hmapx_destroy(&mirror->srcs);
2856 hmapx_destroy(&mirror->dsts);
2857 free(mirror->vlans);
2859 ofproto->mirrors[mirror->idx] = NULL;
2863 mirror_update_dups(ofproto);
2865 ofproto->has_mirrors = false;
2866 for (i = 0; i < MAX_MIRRORS; i++) {
2867 if (ofproto->mirrors[i]) {
2868 ofproto->has_mirrors = true;
2875 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2876 uint64_t *packets, uint64_t *bytes)
2878 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2879 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2882 *packets = *bytes = UINT64_MAX;
2886 *packets = mirror->packet_count;
2887 *bytes = mirror->byte_count;
2893 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2895 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2896 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2897 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2903 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2905 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2906 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2907 return bundle && bundle->mirror_out != 0;
2911 forward_bpdu_changed(struct ofproto *ofproto_)
2913 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2914 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2918 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2921 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2922 mac_learning_set_idle_time(ofproto->ml, idle_time);
2923 mac_learning_set_max_entries(ofproto->ml, max_entries);
2928 static struct ofport_dpif *
2929 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
2931 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2932 return ofport ? ofport_dpif_cast(ofport) : NULL;
2935 static struct ofport_dpif *
2936 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
2938 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2939 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2943 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2944 struct ofproto_port *ofproto_port,
2945 struct dpif_port *dpif_port)
2947 ofproto_port->name = dpif_port->name;
2948 ofproto_port->type = dpif_port->type;
2949 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2952 static struct ofport_dpif *
2953 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
2955 const struct ofproto_dpif *ofproto;
2958 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
2963 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2964 struct ofport *ofport;
2966 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
2967 if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) {
2968 return ofport_dpif_cast(ofport);
2975 port_run_fast(struct ofport_dpif *ofport)
2977 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2978 struct ofpbuf packet;
2980 ofpbuf_init(&packet, 0);
2981 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2982 send_packet(ofport, &packet);
2983 ofpbuf_uninit(&packet);
2988 port_run(struct ofport_dpif *ofport)
2990 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2991 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2992 bool enable = netdev_get_carrier(ofport->up.netdev);
2994 ofport->carrier_seq = carrier_seq;
2996 port_run_fast(ofport);
2998 if (ofport->tnl_port
2999 && tnl_port_reconfigure(&ofport->up, ofport->odp_port,
3000 &ofport->tnl_port)) {
3001 ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true;
3005 int cfm_opup = cfm_get_opup(ofport->cfm);
3007 cfm_run(ofport->cfm);
3008 enable = enable && !cfm_get_fault(ofport->cfm);
3010 if (cfm_opup >= 0) {
3011 enable = enable && cfm_opup;
3015 if (ofport->bundle) {
3016 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3017 if (carrier_changed) {
3018 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3022 if (ofport->may_enable != enable) {
3023 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3025 if (ofproto->has_bundle_action) {
3026 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3030 ofport->may_enable = enable;
3034 port_wait(struct ofport_dpif *ofport)
3037 cfm_wait(ofport->cfm);
3042 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3043 struct ofproto_port *ofproto_port)
3045 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3046 struct dpif_port dpif_port;
3049 if (sset_contains(&ofproto->ghost_ports, devname)) {
3050 const char *type = netdev_get_type_from_name(devname);
3052 /* We may be called before ofproto->up.port_by_name is populated with
3053 * the appropriate ofport. For this reason, we must get the name and
3054 * type from the netdev layer directly. */
3056 const struct ofport *ofport;
3058 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3059 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3060 ofproto_port->name = xstrdup(devname);
3061 ofproto_port->type = xstrdup(type);
3067 if (!sset_contains(&ofproto->ports, devname)) {
3070 error = dpif_port_query_by_name(ofproto->backer->dpif,
3071 devname, &dpif_port);
3073 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3079 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3081 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3082 const char *dp_port_name = netdev_vport_get_dpif_port(netdev);
3083 const char *devname = netdev_get_name(netdev);
3085 if (netdev_vport_is_patch(netdev)) {
3086 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3090 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3091 uint32_t port_no = UINT32_MAX;
3094 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3098 if (netdev_get_tunnel_config(netdev)) {
3099 simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no);
3103 if (netdev_get_tunnel_config(netdev)) {
3104 sset_add(&ofproto->ghost_ports, devname);
3106 sset_add(&ofproto->ports, devname);
3112 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
3114 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3115 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3122 sset_find_and_delete(&ofproto->ghost_ports,
3123 netdev_get_name(ofport->up.netdev));
3124 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3125 if (!ofport->tnl_port) {
3126 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3128 /* The caller is going to close ofport->up.netdev. If this is a
3129 * bonded port, then the bond is using that netdev, so remove it
3130 * from the bond. The client will need to reconfigure everything
3131 * after deleting ports, so then the slave will get re-added. */
3132 bundle_remove(&ofport->up);
3139 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3141 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3144 error = netdev_get_stats(ofport->up.netdev, stats);
3146 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3147 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3149 /* ofproto->stats.tx_packets represents packets that we created
3150 * internally and sent to some port (e.g. packets sent with
3151 * send_packet()). Account for them as if they had come from
3152 * OFPP_LOCAL and got forwarded. */
3154 if (stats->rx_packets != UINT64_MAX) {
3155 stats->rx_packets += ofproto->stats.tx_packets;
3158 if (stats->rx_bytes != UINT64_MAX) {
3159 stats->rx_bytes += ofproto->stats.tx_bytes;
3162 /* ofproto->stats.rx_packets represents packets that were received on
3163 * some port and we processed internally and dropped (e.g. STP).
3164 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3166 if (stats->tx_packets != UINT64_MAX) {
3167 stats->tx_packets += ofproto->stats.rx_packets;
3170 if (stats->tx_bytes != UINT64_MAX) {
3171 stats->tx_bytes += ofproto->stats.rx_bytes;
3178 /* Account packets for LOCAL port. */
3180 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
3181 size_t tx_size, size_t rx_size)
3183 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3186 ofproto->stats.rx_packets++;
3187 ofproto->stats.rx_bytes += rx_size;
3190 ofproto->stats.tx_packets++;
3191 ofproto->stats.tx_bytes += tx_size;
3195 struct port_dump_state {
3200 struct ofproto_port port;
3205 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3207 *statep = xzalloc(sizeof(struct port_dump_state));
3212 port_dump_next(const struct ofproto *ofproto_, void *state_,
3213 struct ofproto_port *port)
3215 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3216 struct port_dump_state *state = state_;
3217 const struct sset *sset;
3218 struct sset_node *node;
3220 if (state->has_port) {
3221 ofproto_port_destroy(&state->port);
3222 state->has_port = false;
3224 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3225 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3228 error = port_query_by_name(ofproto_, node->name, &state->port);
3230 *port = state->port;
3231 state->has_port = true;
3233 } else if (error != ENODEV) {
3238 if (!state->ghost) {
3239 state->ghost = true;
3242 return port_dump_next(ofproto_, state_, port);
3249 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3251 struct port_dump_state *state = state_;
3253 if (state->has_port) {
3254 ofproto_port_destroy(&state->port);
3261 port_poll(const struct ofproto *ofproto_, char **devnamep)
3263 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3265 if (ofproto->port_poll_errno) {
3266 int error = ofproto->port_poll_errno;
3267 ofproto->port_poll_errno = 0;
3271 if (sset_is_empty(&ofproto->port_poll_set)) {
3275 *devnamep = sset_pop(&ofproto->port_poll_set);
3280 port_poll_wait(const struct ofproto *ofproto_)
3282 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3283 dpif_port_poll_wait(ofproto->backer->dpif);
3287 port_is_lacp_current(const struct ofport *ofport_)
3289 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3290 return (ofport->bundle && ofport->bundle->lacp
3291 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3295 /* Upcall handling. */
3297 /* Flow miss batching.
3299 * Some dpifs implement operations faster when you hand them off in a batch.
3300 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3301 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3302 * more packets, plus possibly installing the flow in the dpif.
3304 * So far we only batch the operations that affect flow setup time the most.
3305 * It's possible to batch more than that, but the benefit might be minimal. */
3307 struct hmap_node hmap_node;
3308 struct ofproto_dpif *ofproto;
3310 enum odp_key_fitness key_fitness;
3311 const struct nlattr *key;
3313 struct initial_vals initial_vals;
3314 struct list packets;
3315 enum dpif_upcall_type upcall_type;
3316 uint32_t odp_in_port;
3319 struct flow_miss_op {
3320 struct dpif_op dpif_op;
3321 void *garbage; /* Pointer to pass to free(), NULL if none. */
3322 uint64_t stub[1024 / 8]; /* Temporary buffer. */
3325 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3326 * OpenFlow controller as necessary according to their individual
3327 * configurations. */
3329 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3330 const struct flow *flow)
3332 struct ofputil_packet_in pin;
3334 pin.packet = packet->data;
3335 pin.packet_len = packet->size;
3336 pin.reason = OFPR_NO_MATCH;
3337 pin.controller_id = 0;
3342 pin.send_len = 0; /* not used for flow table misses */
3344 flow_get_metadata(flow, &pin.fmd);
3346 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3349 static enum slow_path_reason
3350 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3351 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
3355 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3357 cfm_process_heartbeat(ofport->cfm, packet);
3360 } else if (ofport->bundle && ofport->bundle->lacp
3361 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3363 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3366 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3368 stp_process_packet(ofport, packet);
3376 static struct flow_miss *
3377 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3378 const struct flow *flow, uint32_t hash)
3380 struct flow_miss *miss;
3382 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3383 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3391 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3392 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3393 * 'miss' is associated with a subfacet the caller must also initialize the
3394 * returned op->subfacet, and if anything needs to be freed after processing
3395 * the op, the caller must initialize op->garbage also. */
3397 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3398 struct flow_miss_op *op)
3400 if (miss->flow.vlan_tci != miss->initial_vals.vlan_tci) {
3401 /* This packet was received on a VLAN splinter port. We
3402 * added a VLAN to the packet to make the packet resemble
3403 * the flow, but the actions were composed assuming that
3404 * the packet contained no VLAN. So, we must remove the
3405 * VLAN header from the packet before trying to execute the
3407 eth_pop_vlan(packet);
3411 op->dpif_op.type = DPIF_OP_EXECUTE;
3412 op->dpif_op.u.execute.key = miss->key;
3413 op->dpif_op.u.execute.key_len = miss->key_len;
3414 op->dpif_op.u.execute.packet = packet;
3417 /* Helper for handle_flow_miss_without_facet() and
3418 * handle_flow_miss_with_facet(). */
3420 handle_flow_miss_common(struct rule_dpif *rule,
3421 struct ofpbuf *packet, const struct flow *flow)
3423 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3425 ofproto->n_matches++;
3427 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3429 * Extra-special case for fail-open mode.
3431 * We are in fail-open mode and the packet matched the fail-open
3432 * rule, but we are connected to a controller too. We should send
3433 * the packet up to the controller in the hope that it will try to
3434 * set up a flow and thereby allow us to exit fail-open.
3436 * See the top-level comment in fail-open.c for more information.
3438 send_packet_in_miss(ofproto, packet, flow);
3442 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3443 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3444 * installing a datapath flow. The answer is usually "yes" (a return value of
3445 * true). However, for short flows the cost of bookkeeping is much higher than
3446 * the benefits, so when the datapath holds a large number of flows we impose
3447 * some heuristics to decide which flows are likely to be worth tracking. */
3449 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3450 struct flow_miss *miss, uint32_t hash)
3452 if (!ofproto->governor) {
3455 n_subfacets = hmap_count(&ofproto->subfacets);
3456 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3460 ofproto->governor = governor_create(ofproto->up.name);
3463 return governor_should_install_flow(ofproto->governor, hash,
3464 list_size(&miss->packets));
3467 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3468 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3469 * increment '*n_ops'. */
3471 handle_flow_miss_without_facet(struct flow_miss *miss,
3472 struct rule_dpif *rule,
3473 struct flow_miss_op *ops, size_t *n_ops)
3475 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3476 long long int now = time_msec();
3477 struct action_xlate_ctx ctx;
3478 struct ofpbuf *packet;
3480 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3481 struct flow_miss_op *op = &ops[*n_ops];
3482 struct dpif_flow_stats stats;
3483 struct ofpbuf odp_actions;
3485 COVERAGE_INC(facet_suppress);
3487 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3489 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3490 rule_credit_stats(rule, &stats);
3492 action_xlate_ctx_init(&ctx, ofproto, &miss->flow,
3493 &miss->initial_vals, rule, 0, packet);
3494 ctx.resubmit_stats = &stats;
3495 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
3498 if (odp_actions.size) {
3499 struct dpif_execute *execute = &op->dpif_op.u.execute;
3501 init_flow_miss_execute_op(miss, packet, op);
3502 execute->actions = odp_actions.data;
3503 execute->actions_len = odp_actions.size;
3504 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3508 ofpbuf_uninit(&odp_actions);
3513 /* Handles 'miss', which matches 'facet'. May add any required datapath
3514 * operations to 'ops', incrementing '*n_ops' for each new op.
3516 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3517 * This is really important only for new facets: if we just called time_msec()
3518 * here, then the new subfacet or its packets could look (occasionally) as
3519 * though it was used some time after the facet was used. That can make a
3520 * one-packet flow look like it has a nonzero duration, which looks odd in
3521 * e.g. NetFlow statistics. */
3523 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3525 struct flow_miss_op *ops, size_t *n_ops)
3527 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3528 enum subfacet_path want_path;
3529 struct subfacet *subfacet;
3530 struct ofpbuf *packet;
3532 subfacet = subfacet_create(facet, miss, now);
3534 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3535 struct flow_miss_op *op = &ops[*n_ops];
3536 struct dpif_flow_stats stats;
3537 struct ofpbuf odp_actions;
3539 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3541 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3542 if (!subfacet->actions || subfacet->slow) {
3543 subfacet_make_actions(subfacet, packet, &odp_actions);
3546 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3547 subfacet_update_stats(subfacet, &stats);
3549 if (subfacet->actions_len) {
3550 struct dpif_execute *execute = &op->dpif_op.u.execute;
3552 init_flow_miss_execute_op(miss, packet, op);
3553 if (!subfacet->slow) {
3554 execute->actions = subfacet->actions;
3555 execute->actions_len = subfacet->actions_len;
3556 ofpbuf_uninit(&odp_actions);
3558 execute->actions = odp_actions.data;
3559 execute->actions_len = odp_actions.size;
3560 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3565 ofpbuf_uninit(&odp_actions);
3569 want_path = subfacet_want_path(subfacet->slow);
3570 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3571 struct flow_miss_op *op = &ops[(*n_ops)++];
3572 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3574 subfacet->path = want_path;
3577 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3578 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3579 put->key = miss->key;
3580 put->key_len = miss->key_len;
3581 if (want_path == SF_FAST_PATH) {
3582 put->actions = subfacet->actions;
3583 put->actions_len = subfacet->actions_len;
3585 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
3586 op->stub, sizeof op->stub,
3587 &put->actions, &put->actions_len);
3593 /* Handles flow miss 'miss'. May add any required datapath operations
3594 * to 'ops', incrementing '*n_ops' for each new op. */
3596 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3599 struct ofproto_dpif *ofproto = miss->ofproto;
3600 struct facet *facet;
3604 /* The caller must ensure that miss->hmap_node.hash contains
3605 * flow_hash(miss->flow, 0). */
3606 hash = miss->hmap_node.hash;
3608 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3610 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
3612 if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
3613 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
3617 facet = facet_create(rule, &miss->flow, hash);
3622 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3625 static struct drop_key *
3626 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3629 struct drop_key *drop_key;
3631 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3632 &backer->drop_keys) {
3633 if (drop_key->key_len == key_len
3634 && !memcmp(drop_key->key, key, key_len)) {
3642 drop_key_clear(struct dpif_backer *backer)
3644 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3645 struct drop_key *drop_key, *next;
3647 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3650 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3652 if (error && !VLOG_DROP_WARN(&rl)) {
3653 struct ds ds = DS_EMPTY_INITIALIZER;
3654 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3655 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3660 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3661 free(drop_key->key);
3666 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3667 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3668 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3669 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3670 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3671 * 'packet' ingressed.
3673 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3674 * 'flow''s in_port to OFPP_NONE.
3676 * This function does post-processing on data returned from
3677 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3678 * of the upcall processing logic. In particular, if the extracted in_port is
3679 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3680 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3681 * a VLAN header onto 'packet' (if it is nonnull).
3683 * Optionally, if 'initial_vals' is nonnull, sets 'initial_vals->vlan_tci'
3684 * to the VLAN TCI with which the packet was really received, that is, the
3685 * actual VLAN TCI extracted by odp_flow_key_to_flow(). (This differs from
3686 * the value returned in flow->vlan_tci only for packets received on
3687 * VLAN splinters.) Also, if received on an IP tunnel, sets
3688 * 'initial_vals->tunnel_ip_tos' to the tunnel's IP TOS.
3690 * Similarly, this function also includes some logic to help with tunnels. It
3691 * may modify 'flow' as necessary to make the tunneling implementation
3692 * transparent to the upcall processing logic.
3694 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3695 * or some other positive errno if there are other problems. */
3697 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3698 const struct nlattr *key, size_t key_len,
3699 struct flow *flow, enum odp_key_fitness *fitnessp,
3700 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3701 struct initial_vals *initial_vals)
3703 const struct ofport_dpif *port;
3704 enum odp_key_fitness fitness;
3707 fitness = odp_flow_key_to_flow(key, key_len, flow);
3708 if (fitness == ODP_FIT_ERROR) {
3714 initial_vals->vlan_tci = flow->vlan_tci;
3715 initial_vals->tunnel_ip_tos = flow->tunnel.ip_tos;
3719 *odp_in_port = flow->in_port;
3722 if (tnl_port_should_receive(flow)) {
3723 const struct ofport *ofport = tnl_port_receive(flow);
3725 flow->in_port = OFPP_NONE;
3728 port = ofport_dpif_cast(ofport);
3730 /* We can't reproduce 'key' from 'flow'. */
3731 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3733 /* XXX: Since the tunnel module is not scoped per backer, it's
3734 * theoretically possible that we'll receive an ofport belonging to an
3735 * entirely different datapath. In practice, this can't happen because
3736 * no platforms has two separate datapaths which each support
3738 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3740 port = odp_port_to_ofport(backer, flow->in_port);
3742 flow->in_port = OFPP_NONE;
3746 flow->in_port = port->up.ofp_port;
3747 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3749 /* Make the packet resemble the flow, so that it gets sent to
3750 * an OpenFlow controller properly, so that it looks correct
3751 * for sFlow, and so that flow_extract() will get the correct
3752 * vlan_tci if it is called on 'packet'.
3754 * The allocated space inside 'packet' probably also contains
3755 * 'key', that is, both 'packet' and 'key' are probably part of
3756 * a struct dpif_upcall (see the large comment on that
3757 * structure definition), so pushing data on 'packet' is in
3758 * general not a good idea since it could overwrite 'key' or
3759 * free it as a side effect. However, it's OK in this special
3760 * case because we know that 'packet' is inside a Netlink
3761 * attribute: pushing 4 bytes will just overwrite the 4-byte
3762 * "struct nlattr", which is fine since we don't need that
3763 * header anymore. */
3764 eth_push_vlan(packet, flow->vlan_tci);
3766 /* We can't reproduce 'key' from 'flow'. */
3767 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3773 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3778 *fitnessp = fitness;
3784 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3787 struct dpif_upcall *upcall;
3788 struct flow_miss *miss;
3789 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3790 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3791 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3801 /* Construct the to-do list.
3803 * This just amounts to extracting the flow from each packet and sticking
3804 * the packets that have the same flow in the same "flow_miss" structure so
3805 * that we can process them together. */
3808 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3809 struct flow_miss *miss = &misses[n_misses];
3810 struct flow_miss *existing_miss;
3811 struct ofproto_dpif *ofproto;
3812 uint32_t odp_in_port;
3817 error = ofproto_receive(backer, upcall->packet, upcall->key,
3818 upcall->key_len, &flow, &miss->key_fitness,
3819 &ofproto, &odp_in_port, &miss->initial_vals);
3820 if (error == ENODEV) {
3821 struct drop_key *drop_key;
3823 /* Received packet on port for which we couldn't associate
3824 * an ofproto. This can happen if a port is removed while
3825 * traffic is being received. Print a rate-limited message
3826 * in case it happens frequently. Install a drop flow so
3827 * that future packets of the flow are inexpensively dropped
3829 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
3832 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
3834 drop_key = xmalloc(sizeof *drop_key);
3835 drop_key->key = xmemdup(upcall->key, upcall->key_len);
3836 drop_key->key_len = upcall->key_len;
3838 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3839 hash_bytes(drop_key->key, drop_key->key_len, 0));
3840 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3841 drop_key->key, drop_key->key_len, NULL, 0, NULL);
3848 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3849 &flow.tunnel, flow.in_port, &miss->flow);
3851 /* Add other packets to a to-do list. */
3852 hash = flow_hash(&miss->flow, 0);
3853 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
3854 if (!existing_miss) {
3855 hmap_insert(&todo, &miss->hmap_node, hash);
3856 miss->ofproto = ofproto;
3857 miss->key = upcall->key;
3858 miss->key_len = upcall->key_len;
3859 miss->upcall_type = upcall->type;
3860 miss->odp_in_port = odp_in_port;
3861 list_init(&miss->packets);
3865 miss = existing_miss;
3867 list_push_back(&miss->packets, &upcall->packet->list_node);
3870 /* Process each element in the to-do list, constructing the set of
3871 * operations to batch. */
3873 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3874 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3876 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3878 /* Execute batch. */
3879 for (i = 0; i < n_ops; i++) {
3880 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3882 dpif_operate(backer->dpif, dpif_ops, n_ops);
3885 for (i = 0; i < n_ops; i++) {
3886 free(flow_miss_ops[i].garbage);
3888 hmap_destroy(&todo);
3891 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL }
3892 classify_upcall(const struct dpif_upcall *upcall)
3894 union user_action_cookie cookie;
3896 /* First look at the upcall type. */
3897 switch (upcall->type) {
3898 case DPIF_UC_ACTION:
3904 case DPIF_N_UC_TYPES:
3906 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3910 /* "action" upcalls need a closer look. */
3911 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3912 switch (cookie.type) {
3913 case USER_ACTION_COOKIE_SFLOW:
3914 return SFLOW_UPCALL;
3916 case USER_ACTION_COOKIE_SLOW_PATH:
3919 case USER_ACTION_COOKIE_UNSPEC:
3921 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
3927 handle_sflow_upcall(struct dpif_backer *backer,
3928 const struct dpif_upcall *upcall)
3930 struct ofproto_dpif *ofproto;
3931 union user_action_cookie cookie;
3933 uint32_t odp_in_port;
3935 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3936 &flow, NULL, &ofproto, &odp_in_port, NULL)
3937 || !ofproto->sflow) {
3941 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3942 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3943 odp_in_port, &cookie);
3947 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
3949 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3950 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3951 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3956 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
3959 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3960 struct dpif_upcall *upcall = &misses[n_misses];
3961 struct ofpbuf *buf = &miss_bufs[n_misses];
3964 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3965 sizeof miss_buf_stubs[n_misses]);
3966 error = dpif_recv(backer->dpif, upcall, buf);
3972 switch (classify_upcall(upcall)) {
3974 /* Handle it later. */
3979 handle_sflow_upcall(backer, upcall);
3989 /* Handle deferred MISS_UPCALL processing. */
3990 handle_miss_upcalls(backer, misses, n_misses);
3991 for (i = 0; i < n_misses; i++) {
3992 ofpbuf_uninit(&miss_bufs[i]);
3998 /* Flow expiration. */
4000 static int subfacet_max_idle(const struct ofproto_dpif *);
4001 static void update_stats(struct dpif_backer *);
4002 static void rule_expire(struct rule_dpif *);
4003 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
4005 /* This function is called periodically by run(). Its job is to collect
4006 * updates for the flows that have been installed into the datapath, most
4007 * importantly when they last were used, and then use that information to
4008 * expire flows that have not been used recently.
4010 * Returns the number of milliseconds after which it should be called again. */
4012 expire(struct dpif_backer *backer)
4014 struct ofproto_dpif *ofproto;
4015 int max_idle = INT32_MAX;
4017 /* Periodically clear out the drop keys in an effort to keep them
4018 * relatively few. */
4019 drop_key_clear(backer);
4021 /* Update stats for each flow in the backer. */
4022 update_stats(backer);
4024 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4025 struct rule *rule, *next_rule;
4028 if (ofproto->backer != backer) {
4032 /* Expire subfacets that have been idle too long. */
4033 dp_max_idle = subfacet_max_idle(ofproto);
4034 expire_subfacets(ofproto, dp_max_idle);
4036 max_idle = MIN(max_idle, dp_max_idle);
4038 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4040 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4041 &ofproto->up.expirable) {
4042 rule_expire(rule_dpif_cast(rule));
4045 /* All outstanding data in existing flows has been accounted, so it's a
4046 * good time to do bond rebalancing. */
4047 if (ofproto->has_bonded_bundles) {
4048 struct ofbundle *bundle;
4050 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4052 bond_rebalance(bundle->bond, &backer->revalidate_set);
4058 return MIN(max_idle, 1000);
4061 /* Updates flow table statistics given that the datapath just reported 'stats'
4062 * as 'subfacet''s statistics. */
4064 update_subfacet_stats(struct subfacet *subfacet,
4065 const struct dpif_flow_stats *stats)
4067 struct facet *facet = subfacet->facet;
4069 if (stats->n_packets >= subfacet->dp_packet_count) {
4070 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
4071 facet->packet_count += extra;
4073 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4076 if (stats->n_bytes >= subfacet->dp_byte_count) {
4077 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
4079 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4082 subfacet->dp_packet_count = stats->n_packets;
4083 subfacet->dp_byte_count = stats->n_bytes;
4085 facet->tcp_flags |= stats->tcp_flags;
4087 subfacet_update_time(subfacet, stats->used);
4088 if (facet->accounted_bytes < facet->byte_count) {
4090 facet_account(facet);
4091 facet->accounted_bytes = facet->byte_count;
4093 facet_push_stats(facet);
4096 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4097 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4099 delete_unexpected_flow(struct ofproto_dpif *ofproto,
4100 const struct nlattr *key, size_t key_len)
4102 if (!VLOG_DROP_WARN(&rl)) {
4106 odp_flow_key_format(key, key_len, &s);
4107 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
4111 COVERAGE_INC(facet_unexpected);
4112 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
4115 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4117 * This function also pushes statistics updates to rules which each facet
4118 * resubmits into. Generally these statistics will be accurate. However, if a
4119 * facet changes the rule it resubmits into at some time in between
4120 * update_stats() runs, it is possible that statistics accrued to the
4121 * old rule will be incorrectly attributed to the new rule. This could be
4122 * avoided by calling update_stats() whenever rules are created or
4123 * deleted. However, the performance impact of making so many calls to the
4124 * datapath do not justify the benefit of having perfectly accurate statistics.
4127 update_stats(struct dpif_backer *backer)
4129 const struct dpif_flow_stats *stats;
4130 struct dpif_flow_dump dump;
4131 const struct nlattr *key;
4134 dpif_flow_dump_start(&dump, backer->dpif);
4135 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4137 struct subfacet *subfacet;
4138 struct ofproto_dpif *ofproto;
4139 struct ofport_dpif *ofport;
4142 if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto,
4147 ofport = get_ofp_port(ofproto, flow.in_port);
4148 if (ofport && ofport->tnl_port) {
4149 netdev_vport_inc_rx(ofport->up.netdev, stats);
4152 key_hash = odp_flow_key_hash(key, key_len);
4153 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
4154 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4156 update_subfacet_stats(subfacet, stats);
4160 /* Stats are updated per-packet. */
4163 case SF_NOT_INSTALLED:
4165 delete_unexpected_flow(ofproto, key, key_len);
4169 dpif_flow_dump_done(&dump);
4172 /* Calculates and returns the number of milliseconds of idle time after which
4173 * subfacets should expire from the datapath. When a subfacet expires, we fold
4174 * its statistics into its facet, and when a facet's last subfacet expires, we
4175 * fold its statistic into its rule. */
4177 subfacet_max_idle(const struct ofproto_dpif *ofproto)
4180 * Idle time histogram.
4182 * Most of the time a switch has a relatively small number of subfacets.
4183 * When this is the case we might as well keep statistics for all of them
4184 * in userspace and to cache them in the kernel datapath for performance as
4187 * As the number of subfacets increases, the memory required to maintain
4188 * statistics about them in userspace and in the kernel becomes
4189 * significant. However, with a large number of subfacets it is likely
4190 * that only a few of them are "heavy hitters" that consume a large amount
4191 * of bandwidth. At this point, only heavy hitters are worth caching in
4192 * the kernel and maintaining in userspaces; other subfacets we can
4195 * The technique used to compute the idle time is to build a histogram with
4196 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4197 * that is installed in the kernel gets dropped in the appropriate bucket.
4198 * After the histogram has been built, we compute the cutoff so that only
4199 * the most-recently-used 1% of subfacets (but at least
4200 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
4201 * the most-recently-used bucket of subfacets is kept, so actually an
4202 * arbitrary number of subfacets can be kept in any given expiration run
4203 * (though the next run will delete most of those unless they receive
4206 * This requires a second pass through the subfacets, in addition to the
4207 * pass made by update_stats(), because the former function never looks at
4208 * uninstallable subfacets.
4210 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4211 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4212 int buckets[N_BUCKETS] = { 0 };
4213 int total, subtotal, bucket;
4214 struct subfacet *subfacet;
4218 total = hmap_count(&ofproto->subfacets);
4219 if (total <= ofproto->up.flow_eviction_threshold) {
4220 return N_BUCKETS * BUCKET_WIDTH;
4223 /* Build histogram. */
4225 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
4226 long long int idle = now - subfacet->used;
4227 int bucket = (idle <= 0 ? 0
4228 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4229 : (unsigned int) idle / BUCKET_WIDTH);
4233 /* Find the first bucket whose flows should be expired. */
4234 subtotal = bucket = 0;
4236 subtotal += buckets[bucket++];
4237 } while (bucket < N_BUCKETS &&
4238 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
4240 if (VLOG_IS_DBG_ENABLED()) {
4244 ds_put_cstr(&s, "keep");
4245 for (i = 0; i < N_BUCKETS; i++) {
4247 ds_put_cstr(&s, ", drop");
4250 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4253 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
4257 return bucket * BUCKET_WIDTH;
4261 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
4263 /* Cutoff time for most flows. */
4264 long long int normal_cutoff = time_msec() - dp_max_idle;
4266 /* We really want to keep flows for special protocols around, so use a more
4267 * conservative cutoff. */
4268 long long int special_cutoff = time_msec() - 10000;
4270 struct subfacet *subfacet, *next_subfacet;
4271 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4275 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4276 &ofproto->subfacets) {
4277 long long int cutoff;
4279 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)
4282 if (subfacet->used < cutoff) {
4283 if (subfacet->path != SF_NOT_INSTALLED) {
4284 batch[n_batch++] = subfacet;
4285 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4286 subfacet_destroy_batch(ofproto, batch, n_batch);
4290 subfacet_destroy(subfacet);
4296 subfacet_destroy_batch(ofproto, batch, n_batch);
4300 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4301 * then delete it entirely. */
4303 rule_expire(struct rule_dpif *rule)
4305 struct facet *facet, *next_facet;
4309 if (rule->up.pending) {
4310 /* We'll have to expire it later. */
4314 /* Has 'rule' expired? */
4316 if (rule->up.hard_timeout
4317 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4318 reason = OFPRR_HARD_TIMEOUT;
4319 } else if (rule->up.idle_timeout
4320 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4321 reason = OFPRR_IDLE_TIMEOUT;
4326 COVERAGE_INC(ofproto_dpif_expired);
4328 /* Update stats. (This is a no-op if the rule expired due to an idle
4329 * timeout, because that only happens when the rule has no facets left.) */
4330 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4331 facet_remove(facet);
4334 /* Get rid of the rule. */
4335 ofproto_rule_expire(&rule->up, reason);
4340 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
4342 * The caller must already have determined that no facet with an identical
4343 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
4344 * the ofproto's classifier table.
4346 * 'hash' must be the return value of flow_hash(flow, 0).
4348 * The facet will initially have no subfacets. The caller should create (at
4349 * least) one subfacet with subfacet_create(). */
4350 static struct facet *
4351 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
4353 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4354 struct facet *facet;
4356 facet = xzalloc(sizeof *facet);
4357 facet->used = time_msec();
4358 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4359 list_push_back(&rule->facets, &facet->list_node);
4361 facet->flow = *flow;
4362 list_init(&facet->subfacets);
4363 netflow_flow_init(&facet->nf_flow);
4364 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4370 facet_free(struct facet *facet)
4375 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4376 * 'packet', which arrived on 'in_port'. */
4378 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4379 const struct nlattr *odp_actions, size_t actions_len,
4380 struct ofpbuf *packet)
4382 struct odputil_keybuf keybuf;
4386 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4387 odp_flow_key_from_flow(&key, flow,
4388 ofp_port_to_odp_port(ofproto, flow->in_port));
4390 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4391 odp_actions, actions_len, packet);
4395 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4397 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4398 * rule's statistics, via subfacet_uninstall().
4400 * - Removes 'facet' from its rule and from ofproto->facets.
4403 facet_remove(struct facet *facet)
4405 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4406 struct subfacet *subfacet, *next_subfacet;
4408 ovs_assert(!list_is_empty(&facet->subfacets));
4410 /* First uninstall all of the subfacets to get final statistics. */
4411 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4412 subfacet_uninstall(subfacet);
4415 /* Flush the final stats to the rule.
4417 * This might require us to have at least one subfacet around so that we
4418 * can use its actions for accounting in facet_account(), which is why we
4419 * have uninstalled but not yet destroyed the subfacets. */
4420 facet_flush_stats(facet);
4422 /* Now we're really all done so destroy everything. */
4423 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4424 &facet->subfacets) {
4425 subfacet_destroy__(subfacet);
4427 hmap_remove(&ofproto->facets, &facet->hmap_node);
4428 list_remove(&facet->list_node);
4432 /* Feed information from 'facet' back into the learning table to keep it in
4433 * sync with what is actually flowing through the datapath. */
4435 facet_learn(struct facet *facet)
4437 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4438 struct subfacet *subfacet= CONTAINER_OF(list_front(&facet->subfacets),
4439 struct subfacet, list_node);
4440 struct action_xlate_ctx ctx;
4442 if (!facet->has_learn
4443 && !facet->has_normal
4444 && (!facet->has_fin_timeout
4445 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4449 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4450 &subfacet->initial_vals,
4451 facet->rule, facet->tcp_flags, NULL);
4452 ctx.may_learn = true;
4453 xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
4454 facet->rule->up.ofpacts_len);
4458 facet_account(struct facet *facet)
4460 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4461 struct subfacet *subfacet;
4462 const struct nlattr *a;
4467 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
4470 n_bytes = facet->byte_count - facet->accounted_bytes;
4472 /* This loop feeds byte counters to bond_account() for rebalancing to use
4473 * as a basis. We also need to track the actual VLAN on which the packet
4474 * is going to be sent to ensure that it matches the one passed to
4475 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4478 * We use the actions from an arbitrary subfacet because they should all
4479 * be equally valid for our purpose. */
4480 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
4481 struct subfacet, list_node);
4482 vlan_tci = facet->flow.vlan_tci;
4483 NL_ATTR_FOR_EACH_UNSAFE (a, left,
4484 subfacet->actions, subfacet->actions_len) {
4485 const struct ovs_action_push_vlan *vlan;
4486 struct ofport_dpif *port;
4488 switch (nl_attr_type(a)) {
4489 case OVS_ACTION_ATTR_OUTPUT:
4490 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4491 if (port && port->bundle && port->bundle->bond) {
4492 bond_account(port->bundle->bond, &facet->flow,
4493 vlan_tci_to_vid(vlan_tci), n_bytes);
4497 case OVS_ACTION_ATTR_POP_VLAN:
4498 vlan_tci = htons(0);
4501 case OVS_ACTION_ATTR_PUSH_VLAN:
4502 vlan = nl_attr_get(a);
4503 vlan_tci = vlan->vlan_tci;
4509 /* Returns true if the only action for 'facet' is to send to the controller.
4510 * (We don't report NetFlow expiration messages for such facets because they
4511 * are just part of the control logic for the network, not real traffic). */
4513 facet_is_controller_flow(struct facet *facet)
4516 const struct rule *rule = &facet->rule->up;
4517 const struct ofpact *ofpacts = rule->ofpacts;
4518 size_t ofpacts_len = rule->ofpacts_len;
4520 if (ofpacts_len > 0 &&
4521 ofpacts->type == OFPACT_CONTROLLER &&
4522 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4529 /* Folds all of 'facet''s statistics into its rule. Also updates the
4530 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4531 * 'facet''s statistics in the datapath should have been zeroed and folded into
4532 * its packet and byte counts before this function is called. */
4534 facet_flush_stats(struct facet *facet)
4536 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4537 struct subfacet *subfacet;
4539 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4540 ovs_assert(!subfacet->dp_byte_count);
4541 ovs_assert(!subfacet->dp_packet_count);
4544 facet_push_stats(facet);
4545 if (facet->accounted_bytes < facet->byte_count) {
4546 facet_account(facet);
4547 facet->accounted_bytes = facet->byte_count;
4550 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4551 struct ofexpired expired;
4552 expired.flow = facet->flow;
4553 expired.packet_count = facet->packet_count;
4554 expired.byte_count = facet->byte_count;
4555 expired.used = facet->used;
4556 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4559 facet->rule->packet_count += facet->packet_count;
4560 facet->rule->byte_count += facet->byte_count;
4562 /* Reset counters to prevent double counting if 'facet' ever gets
4564 facet_reset_counters(facet);
4566 netflow_flow_clear(&facet->nf_flow);
4567 facet->tcp_flags = 0;
4570 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4571 * Returns it if found, otherwise a null pointer.
4573 * 'hash' must be the return value of flow_hash(flow, 0).
4575 * The returned facet might need revalidation; use facet_lookup_valid()
4576 * instead if that is important. */
4577 static struct facet *
4578 facet_find(struct ofproto_dpif *ofproto,
4579 const struct flow *flow, uint32_t hash)
4581 struct facet *facet;
4583 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4584 if (flow_equal(flow, &facet->flow)) {
4592 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4593 * Returns it if found, otherwise a null pointer.
4595 * 'hash' must be the return value of flow_hash(flow, 0).
4597 * The returned facet is guaranteed to be valid. */
4598 static struct facet *
4599 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4602 struct facet *facet;
4604 facet = facet_find(ofproto, flow, hash);
4606 && (ofproto->backer->need_revalidate
4607 || tag_set_intersects(&ofproto->backer->revalidate_set,
4609 facet_revalidate(facet);
4611 /* facet_revalidate() may have destroyed 'facet'. */
4612 facet = facet_find(ofproto, flow, hash);
4619 subfacet_path_to_string(enum subfacet_path path)
4622 case SF_NOT_INSTALLED:
4623 return "not installed";
4625 return "in fast path";
4627 return "in slow path";
4633 /* Returns the path in which a subfacet should be installed if its 'slow'
4634 * member has the specified value. */
4635 static enum subfacet_path
4636 subfacet_want_path(enum slow_path_reason slow)
4638 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
4641 /* Returns true if 'subfacet' needs to have its datapath flow updated,
4642 * supposing that its actions have been recalculated as 'want_actions' and that
4643 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
4645 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
4646 const struct ofpbuf *want_actions)
4648 enum subfacet_path want_path = subfacet_want_path(slow);
4649 return (want_path != subfacet->path
4650 || (want_path == SF_FAST_PATH
4651 && (subfacet->actions_len != want_actions->size
4652 || memcmp(subfacet->actions, want_actions->data,
4653 subfacet->actions_len))));
4657 facet_check_consistency(struct facet *facet)
4659 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4661 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4663 uint64_t odp_actions_stub[1024 / 8];
4664 struct ofpbuf odp_actions;
4666 struct rule_dpif *rule;
4667 struct subfacet *subfacet;
4668 bool may_log = false;
4671 /* Check the rule for consistency. */
4672 rule = rule_dpif_lookup(ofproto, &facet->flow);
4673 ok = rule == facet->rule;
4675 may_log = !VLOG_DROP_WARN(&rl);
4680 flow_format(&s, &facet->flow);
4681 ds_put_format(&s, ": facet associated with wrong rule (was "
4682 "table=%"PRIu8",", facet->rule->up.table_id);
4683 cls_rule_format(&facet->rule->up.cr, &s);
4684 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4686 cls_rule_format(&rule->up.cr, &s);
4687 ds_put_char(&s, ')');
4689 VLOG_WARN("%s", ds_cstr(&s));
4694 /* Check the datapath actions for consistency. */
4695 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4696 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4697 enum subfacet_path want_path;
4698 struct action_xlate_ctx ctx;
4701 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4702 &subfacet->initial_vals, rule, 0, NULL);
4703 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
4706 if (subfacet->path == SF_NOT_INSTALLED) {
4707 /* This only happens if the datapath reported an error when we
4708 * tried to install the flow. Don't flag another error here. */
4712 want_path = subfacet_want_path(subfacet->slow);
4713 if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) {
4714 /* The actions for slow-path flows may legitimately vary from one
4715 * packet to the next. We're done. */
4719 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
4723 /* Inconsistency! */
4725 may_log = !VLOG_DROP_WARN(&rl);
4729 /* Rate-limited, skip reporting. */
4734 odp_flow_key_format(subfacet->key, subfacet->key_len, &s);
4736 ds_put_cstr(&s, ": inconsistency in subfacet");
4737 if (want_path != subfacet->path) {
4738 enum odp_key_fitness fitness = subfacet->key_fitness;
4740 ds_put_format(&s, " (%s, fitness=%s)",
4741 subfacet_path_to_string(subfacet->path),
4742 odp_key_fitness_to_string(fitness));
4743 ds_put_format(&s, " (should have been %s)",
4744 subfacet_path_to_string(want_path));
4745 } else if (want_path == SF_FAST_PATH) {
4746 ds_put_cstr(&s, " (actions were: ");
4747 format_odp_actions(&s, subfacet->actions,
4748 subfacet->actions_len);
4749 ds_put_cstr(&s, ") (correct actions: ");
4750 format_odp_actions(&s, odp_actions.data, odp_actions.size);
4751 ds_put_char(&s, ')');
4753 ds_put_cstr(&s, " (actions: ");
4754 format_odp_actions(&s, subfacet->actions,
4755 subfacet->actions_len);
4756 ds_put_char(&s, ')');
4758 VLOG_WARN("%s", ds_cstr(&s));
4761 ofpbuf_uninit(&odp_actions);
4766 /* Re-searches the classifier for 'facet':
4768 * - If the rule found is different from 'facet''s current rule, moves
4769 * 'facet' to the new rule and recompiles its actions.
4771 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4772 * where it is and recompiles its actions anyway.
4774 * - If any of 'facet''s subfacets correspond to a new flow according to
4775 * ofproto_receive(), 'facet' is removed. */
4777 facet_revalidate(struct facet *facet)
4779 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4781 struct nlattr *odp_actions;
4784 struct actions *new_actions;
4786 struct action_xlate_ctx ctx;
4787 uint64_t odp_actions_stub[1024 / 8];
4788 struct ofpbuf odp_actions;
4790 struct rule_dpif *new_rule;
4791 struct subfacet *subfacet;
4794 COVERAGE_INC(facet_revalidate);
4796 /* Check that child subfacets still correspond to this facet. Tunnel
4797 * configuration changes could cause a subfacet's OpenFlow in_port to
4799 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4800 struct ofproto_dpif *recv_ofproto;
4801 struct flow recv_flow;
4804 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
4805 subfacet->key_len, &recv_flow, NULL,
4806 &recv_ofproto, NULL, NULL);
4808 || recv_ofproto != ofproto
4809 || memcmp(&recv_flow, &facet->flow, sizeof recv_flow)) {
4810 facet_remove(facet);
4815 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
4817 /* Calculate new datapath actions.
4819 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4820 * emit a NetFlow expiration and, if so, we need to have the old state
4821 * around to properly compose it. */
4823 /* If the datapath actions changed or the installability changed,
4824 * then we need to talk to the datapath. */
4827 memset(&ctx, 0, sizeof ctx);
4828 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4829 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4830 enum slow_path_reason slow;
4832 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4833 &subfacet->initial_vals, new_rule, 0, NULL);
4834 xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len,
4837 slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4838 if (subfacet_should_install(subfacet, slow, &odp_actions)) {
4839 struct dpif_flow_stats stats;
4841 subfacet_install(subfacet,
4842 odp_actions.data, odp_actions.size, &stats, slow);
4843 subfacet_update_stats(subfacet, &stats);
4846 new_actions = xcalloc(list_size(&facet->subfacets),
4847 sizeof *new_actions);
4849 new_actions[i].odp_actions = xmemdup(odp_actions.data,
4851 new_actions[i].actions_len = odp_actions.size;
4856 ofpbuf_uninit(&odp_actions);
4859 facet_flush_stats(facet);
4862 /* Update 'facet' now that we've taken care of all the old state. */
4863 facet->tags = ctx.tags;
4864 facet->nf_flow.output_iface = ctx.nf_output_iface;
4865 facet->has_learn = ctx.has_learn;
4866 facet->has_normal = ctx.has_normal;
4867 facet->has_fin_timeout = ctx.has_fin_timeout;
4868 facet->mirrors = ctx.mirrors;
4871 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4872 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4874 if (new_actions && new_actions[i].odp_actions) {
4875 free(subfacet->actions);
4876 subfacet->actions = new_actions[i].odp_actions;
4877 subfacet->actions_len = new_actions[i].actions_len;
4883 if (facet->rule != new_rule) {
4884 COVERAGE_INC(facet_changed_rule);
4885 list_remove(&facet->list_node);
4886 list_push_back(&new_rule->facets, &facet->list_node);
4887 facet->rule = new_rule;
4888 facet->used = new_rule->up.created;
4889 facet->prev_used = facet->used;
4893 /* Updates 'facet''s used time. Caller is responsible for calling
4894 * facet_push_stats() to update the flows which 'facet' resubmits into. */
4896 facet_update_time(struct facet *facet, long long int used)
4898 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4899 if (used > facet->used) {
4901 ofproto_rule_update_used(&facet->rule->up, used);
4902 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
4907 facet_reset_counters(struct facet *facet)
4909 facet->packet_count = 0;
4910 facet->byte_count = 0;
4911 facet->prev_packet_count = 0;
4912 facet->prev_byte_count = 0;
4913 facet->accounted_bytes = 0;
4917 facet_push_stats(struct facet *facet)
4919 struct dpif_flow_stats stats;
4921 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4922 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4923 ovs_assert(facet->used >= facet->prev_used);
4925 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4926 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4927 stats.used = facet->used;
4928 stats.tcp_flags = 0;
4930 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
4931 facet->prev_packet_count = facet->packet_count;
4932 facet->prev_byte_count = facet->byte_count;
4933 facet->prev_used = facet->used;
4935 flow_push_stats(facet, &stats);
4937 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
4938 facet->mirrors, stats.n_packets, stats.n_bytes);
4943 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4945 rule->packet_count += stats->n_packets;
4946 rule->byte_count += stats->n_bytes;
4947 ofproto_rule_update_used(&rule->up, stats->used);
4950 /* Pushes flow statistics to the rules which 'facet->flow' resubmits
4951 * into given 'facet->rule''s actions and mirrors. */
4953 flow_push_stats(struct facet *facet, const struct dpif_flow_stats *stats)
4955 struct rule_dpif *rule = facet->rule;
4956 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4957 struct subfacet *subfacet = CONTAINER_OF(list_front(&facet->subfacets),
4958 struct subfacet, list_node);
4959 struct action_xlate_ctx ctx;
4961 ofproto_rule_update_used(&rule->up, stats->used);
4963 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4964 &subfacet->initial_vals, rule, 0, NULL);
4965 ctx.resubmit_stats = stats;
4966 xlate_actions_for_side_effects(&ctx, rule->up.ofpacts,
4967 rule->up.ofpacts_len);
4972 static struct subfacet *
4973 subfacet_find(struct ofproto_dpif *ofproto,
4974 const struct nlattr *key, size_t key_len, uint32_t key_hash)
4976 struct subfacet *subfacet;
4978 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4979 &ofproto->subfacets) {
4980 if (subfacet->key_len == key_len
4981 && !memcmp(key, subfacet->key, key_len)) {
4989 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4990 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4991 * existing subfacet if there is one, otherwise creates and returns a
4994 * If the returned subfacet is new, then subfacet->actions will be NULL, in
4995 * which case the caller must populate the actions with
4996 * subfacet_make_actions(). */
4997 static struct subfacet *
4998 subfacet_create(struct facet *facet, struct flow_miss *miss,
5001 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5002 enum odp_key_fitness key_fitness = miss->key_fitness;
5003 const struct nlattr *key = miss->key;
5004 size_t key_len = miss->key_len;
5006 struct subfacet *subfacet;
5008 key_hash = odp_flow_key_hash(key, key_len);
5010 if (list_is_empty(&facet->subfacets)) {
5011 subfacet = &facet->one_subfacet;
5013 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
5015 if (subfacet->facet == facet) {
5019 /* This shouldn't happen. */
5020 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
5021 subfacet_destroy(subfacet);
5024 subfacet = xmalloc(sizeof *subfacet);
5027 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
5028 list_push_back(&facet->subfacets, &subfacet->list_node);
5029 subfacet->facet = facet;
5030 subfacet->key_fitness = key_fitness;
5031 subfacet->key = xmemdup(key, key_len);
5032 subfacet->key_len = key_len;
5033 subfacet->used = now;
5034 subfacet->dp_packet_count = 0;
5035 subfacet->dp_byte_count = 0;
5036 subfacet->actions_len = 0;
5037 subfacet->actions = NULL;
5038 subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE
5041 subfacet->path = SF_NOT_INSTALLED;
5042 subfacet->initial_vals = miss->initial_vals;
5043 subfacet->odp_in_port = miss->odp_in_port;
5048 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
5049 * its facet within 'ofproto', and frees it. */
5051 subfacet_destroy__(struct subfacet *subfacet)
5053 struct facet *facet = subfacet->facet;
5054 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5056 subfacet_uninstall(subfacet);
5057 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
5058 list_remove(&subfacet->list_node);
5059 free(subfacet->key);
5060 free(subfacet->actions);
5061 if (subfacet != &facet->one_subfacet) {
5066 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5067 * last remaining subfacet in its facet destroys the facet too. */
5069 subfacet_destroy(struct subfacet *subfacet)
5071 struct facet *facet = subfacet->facet;
5073 if (list_is_singleton(&facet->subfacets)) {
5074 /* facet_remove() needs at least one subfacet (it will remove it). */
5075 facet_remove(facet);
5077 subfacet_destroy__(subfacet);
5082 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
5083 struct subfacet **subfacets, int n)
5085 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5086 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5087 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5090 for (i = 0; i < n; i++) {
5091 ops[i].type = DPIF_OP_FLOW_DEL;
5092 ops[i].u.flow_del.key = subfacets[i]->key;
5093 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
5094 ops[i].u.flow_del.stats = &stats[i];
5098 dpif_operate(ofproto->backer->dpif, opsp, n);
5099 for (i = 0; i < n; i++) {
5100 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5101 subfacets[i]->path = SF_NOT_INSTALLED;
5102 subfacet_destroy(subfacets[i]);
5106 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
5107 * Translates the actions into 'odp_actions', which the caller must have
5108 * initialized and is responsible for uninitializing. */
5110 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
5111 struct ofpbuf *odp_actions)
5113 struct facet *facet = subfacet->facet;
5114 struct rule_dpif *rule = facet->rule;
5115 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5117 struct action_xlate_ctx ctx;
5119 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
5120 &subfacet->initial_vals, rule, 0, packet);
5121 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions);
5122 facet->tags = ctx.tags;
5123 facet->has_learn = ctx.has_learn;
5124 facet->has_normal = ctx.has_normal;
5125 facet->has_fin_timeout = ctx.has_fin_timeout;
5126 facet->nf_flow.output_iface = ctx.nf_output_iface;
5127 facet->mirrors = ctx.mirrors;
5129 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
5130 if (subfacet->actions_len != odp_actions->size
5131 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
5132 free(subfacet->actions);
5133 subfacet->actions_len = odp_actions->size;
5134 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
5138 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5139 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5140 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5141 * since 'subfacet' was last updated.
5143 * Returns 0 if successful, otherwise a positive errno value. */
5145 subfacet_install(struct subfacet *subfacet,
5146 const struct nlattr *actions, size_t actions_len,
5147 struct dpif_flow_stats *stats,
5148 enum slow_path_reason slow)
5150 struct facet *facet = subfacet->facet;
5151 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5152 enum subfacet_path path = subfacet_want_path(slow);
5153 uint64_t slow_path_stub[128 / 8];
5154 enum dpif_flow_put_flags flags;
5157 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5159 flags |= DPIF_FP_ZERO_STATS;
5162 if (path == SF_SLOW_PATH) {
5163 compose_slow_path(ofproto, &facet->flow, slow,
5164 slow_path_stub, sizeof slow_path_stub,
5165 &actions, &actions_len);
5168 ret = dpif_flow_put(ofproto->backer->dpif, flags, subfacet->key,
5169 subfacet->key_len, actions, actions_len, stats);
5172 subfacet_reset_dp_stats(subfacet, stats);
5176 subfacet->path = path;
5182 subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats)
5184 return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len,
5185 stats, subfacet->slow);
5188 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5190 subfacet_uninstall(struct subfacet *subfacet)
5192 if (subfacet->path != SF_NOT_INSTALLED) {
5193 struct rule_dpif *rule = subfacet->facet->rule;
5194 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5195 struct dpif_flow_stats stats;
5198 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5199 subfacet->key_len, &stats);
5200 subfacet_reset_dp_stats(subfacet, &stats);
5202 subfacet_update_stats(subfacet, &stats);
5204 subfacet->path = SF_NOT_INSTALLED;
5206 ovs_assert(subfacet->dp_packet_count == 0);
5207 ovs_assert(subfacet->dp_byte_count == 0);
5211 /* Resets 'subfacet''s datapath statistics counters. This should be called
5212 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5213 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5214 * was reset in the datapath. 'stats' will be modified to include only
5215 * statistics new since 'subfacet' was last updated. */
5217 subfacet_reset_dp_stats(struct subfacet *subfacet,
5218 struct dpif_flow_stats *stats)
5221 && subfacet->dp_packet_count <= stats->n_packets
5222 && subfacet->dp_byte_count <= stats->n_bytes) {
5223 stats->n_packets -= subfacet->dp_packet_count;
5224 stats->n_bytes -= subfacet->dp_byte_count;
5227 subfacet->dp_packet_count = 0;
5228 subfacet->dp_byte_count = 0;
5231 /* Updates 'subfacet''s used time. The caller is responsible for calling
5232 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
5234 subfacet_update_time(struct subfacet *subfacet, long long int used)
5236 if (used > subfacet->used) {
5237 subfacet->used = used;
5238 facet_update_time(subfacet->facet, used);
5242 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5244 * Because of the meaning of a subfacet's counters, it only makes sense to do
5245 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5246 * represents a packet that was sent by hand or if it represents statistics
5247 * that have been cleared out of the datapath. */
5249 subfacet_update_stats(struct subfacet *subfacet,
5250 const struct dpif_flow_stats *stats)
5252 if (stats->n_packets || stats->used > subfacet->used) {
5253 struct facet *facet = subfacet->facet;
5255 subfacet_update_time(subfacet, stats->used);
5256 facet->packet_count += stats->n_packets;
5257 facet->byte_count += stats->n_bytes;
5258 facet->tcp_flags |= stats->tcp_flags;
5259 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
5265 static struct rule_dpif *
5266 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
5268 struct rule_dpif *rule;
5270 rule = rule_dpif_lookup__(ofproto, flow, 0);
5275 return rule_dpif_miss_rule(ofproto, flow);
5278 static struct rule_dpif *
5279 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5282 struct cls_rule *cls_rule;
5283 struct classifier *cls;
5285 if (table_id >= N_TABLES) {
5289 cls = &ofproto->up.tables[table_id].cls;
5290 if (flow->nw_frag & FLOW_NW_FRAG_ANY
5291 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5292 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
5293 * are unavailable. */
5294 struct flow ofpc_normal_flow = *flow;
5295 ofpc_normal_flow.tp_src = htons(0);
5296 ofpc_normal_flow.tp_dst = htons(0);
5297 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
5299 cls_rule = classifier_lookup(cls, flow);
5301 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5304 static struct rule_dpif *
5305 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5307 struct ofport_dpif *port;
5309 port = get_ofp_port(ofproto, flow->in_port);
5311 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5312 return ofproto->miss_rule;
5315 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5316 return ofproto->no_packet_in_rule;
5318 return ofproto->miss_rule;
5322 complete_operation(struct rule_dpif *rule)
5324 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5326 rule_invalidate(rule);
5328 struct dpif_completion *c = xmalloc(sizeof *c);
5329 c->op = rule->up.pending;
5330 list_push_back(&ofproto->completions, &c->list_node);
5332 ofoperation_complete(rule->up.pending, 0);
5336 static struct rule *
5339 struct rule_dpif *rule = xmalloc(sizeof *rule);
5344 rule_dealloc(struct rule *rule_)
5346 struct rule_dpif *rule = rule_dpif_cast(rule_);
5351 rule_construct(struct rule *rule_)
5353 struct rule_dpif *rule = rule_dpif_cast(rule_);
5354 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5355 struct rule_dpif *victim;
5358 rule->packet_count = 0;
5359 rule->byte_count = 0;
5361 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5362 if (victim && !list_is_empty(&victim->facets)) {
5363 struct facet *facet;
5365 rule->facets = victim->facets;
5366 list_moved(&rule->facets);
5367 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5368 /* XXX: We're only clearing our local counters here. It's possible
5369 * that quite a few packets are unaccounted for in the datapath
5370 * statistics. These will be accounted to the new rule instead of
5371 * cleared as required. This could be fixed by clearing out the
5372 * datapath statistics for this facet, but currently it doesn't
5374 facet_reset_counters(facet);
5378 /* Must avoid list_moved() in this case. */
5379 list_init(&rule->facets);
5382 table_id = rule->up.table_id;
5384 rule->tag = victim->tag;
5385 } else if (table_id == 0) {
5390 miniflow_expand(&rule->up.cr.match.flow, &flow);
5391 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5392 ofproto->tables[table_id].basis);
5395 complete_operation(rule);
5400 rule_destruct(struct rule *rule_)
5402 struct rule_dpif *rule = rule_dpif_cast(rule_);
5403 struct facet *facet, *next_facet;
5405 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5406 facet_revalidate(facet);
5409 complete_operation(rule);
5413 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5415 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule_->ofproto);
5416 struct rule_dpif *rule = rule_dpif_cast(rule_);
5417 struct facet *facet;
5419 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5420 facet_push_stats(facet);
5423 /* Start from historical data for 'rule' itself that are no longer tracked
5424 * in facets. This counts, for example, facets that have expired. */
5425 *packets = rule->packet_count;
5426 *bytes = rule->byte_count;
5428 /* Add any statistics that are tracked by facets. This includes
5429 * statistical data recently updated by ofproto_update_stats() as well as
5430 * stats for packets that were executed "by hand" via dpif_execute(). */
5431 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5432 *packets += facet->packet_count;
5433 *bytes += facet->byte_count;
5438 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5439 struct ofpbuf *packet)
5441 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5442 struct initial_vals initial_vals;
5443 struct dpif_flow_stats stats;
5444 struct action_xlate_ctx ctx;
5445 uint64_t odp_actions_stub[1024 / 8];
5446 struct ofpbuf odp_actions;
5448 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5449 rule_credit_stats(rule, &stats);
5451 initial_vals.vlan_tci = flow->vlan_tci;
5452 initial_vals.tunnel_ip_tos = flow->tunnel.ip_tos;
5453 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5454 action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals,
5455 rule, stats.tcp_flags, packet);
5456 ctx.resubmit_stats = &stats;
5457 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
5459 execute_odp_actions(ofproto, flow, odp_actions.data,
5460 odp_actions.size, packet);
5462 ofpbuf_uninit(&odp_actions);
5466 rule_execute(struct rule *rule, const struct flow *flow,
5467 struct ofpbuf *packet)
5469 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5470 ofpbuf_delete(packet);
5475 rule_modify_actions(struct rule *rule_)
5477 struct rule_dpif *rule = rule_dpif_cast(rule_);
5479 complete_operation(rule);
5482 /* Sends 'packet' out 'ofport'.
5483 * May modify 'packet'.
5484 * Returns 0 if successful, otherwise a positive errno value. */
5486 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5488 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5489 uint64_t odp_actions_stub[1024 / 8];
5490 struct ofpbuf key, odp_actions;
5491 struct odputil_keybuf keybuf;
5496 flow_extract(packet, 0, 0, NULL, OFPP_LOCAL, &flow);
5497 if (netdev_vport_is_patch(ofport->up.netdev)) {
5498 struct ofproto_dpif *peer_ofproto;
5499 struct dpif_flow_stats stats;
5500 struct ofport_dpif *peer;
5501 struct rule_dpif *rule;
5503 peer = ofport_get_peer(ofport);
5508 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5509 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5510 netdev_vport_inc_rx(peer->up.netdev, &stats);
5512 flow.in_port = peer->up.ofp_port;
5513 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5514 rule = rule_dpif_lookup(peer_ofproto, &flow);
5515 rule_dpif_execute(rule, &flow, packet);
5520 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5522 if (ofport->tnl_port) {
5523 struct dpif_flow_stats stats;
5525 odp_port = tnl_port_send(ofport->tnl_port, &flow);
5526 if (odp_port == OVSP_NONE) {
5530 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5531 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5532 odp_put_tunnel_action(&flow.tunnel, &odp_actions);
5533 odp_put_skb_mark_action(flow.skb_mark, &odp_actions);
5535 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
5537 if (odp_port != ofport->odp_port) {
5538 eth_pop_vlan(packet);
5539 flow.vlan_tci = htons(0);
5543 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5544 odp_flow_key_from_flow(&key, &flow,
5545 ofp_port_to_odp_port(ofproto, flow.in_port));
5547 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
5549 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
5550 error = dpif_execute(ofproto->backer->dpif,
5552 odp_actions.data, odp_actions.size,
5554 ofpbuf_uninit(&odp_actions);
5557 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
5558 ofproto->up.name, odp_port, strerror(error));
5560 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
5564 /* OpenFlow to datapath action translation. */
5566 static bool may_receive(const struct ofport_dpif *, struct action_xlate_ctx *);
5567 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5568 struct action_xlate_ctx *);
5569 static void xlate_normal(struct action_xlate_ctx *);
5571 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5572 * The action will state 'slow' as the reason that the action is in the slow
5573 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5574 * dump-flows" output to see why a flow is in the slow path.)
5576 * The 'stub_size' bytes in 'stub' will be used to store the action.
5577 * 'stub_size' must be large enough for the action.
5579 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5582 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5583 enum slow_path_reason slow,
5584 uint64_t *stub, size_t stub_size,
5585 const struct nlattr **actionsp, size_t *actions_lenp)
5587 union user_action_cookie cookie;
5590 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5591 cookie.slow_path.unused = 0;
5592 cookie.slow_path.reason = slow;
5594 ofpbuf_use_stack(&buf, stub, stub_size);
5595 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
5596 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5597 odp_put_userspace_action(pid, &cookie, &buf);
5599 put_userspace_action(ofproto, &buf, flow, &cookie);
5601 *actionsp = buf.data;
5602 *actions_lenp = buf.size;
5606 put_userspace_action(const struct ofproto_dpif *ofproto,
5607 struct ofpbuf *odp_actions,
5608 const struct flow *flow,
5609 const union user_action_cookie *cookie)
5613 pid = dpif_port_get_pid(ofproto->backer->dpif,
5614 ofp_port_to_odp_port(ofproto, flow->in_port));
5616 return odp_put_userspace_action(pid, cookie, odp_actions);
5620 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5621 ovs_be16 vlan_tci, uint32_t odp_port,
5622 unsigned int n_outputs, union user_action_cookie *cookie)
5626 cookie->type = USER_ACTION_COOKIE_SFLOW;
5627 cookie->sflow.vlan_tci = vlan_tci;
5629 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5630 * port information") for the interpretation of cookie->output. */
5631 switch (n_outputs) {
5633 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5634 cookie->sflow.output = 0x40000000 | 256;
5638 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5640 cookie->sflow.output = ifindex;
5645 /* 0x80000000 means "multiple output ports. */
5646 cookie->sflow.output = 0x80000000 | n_outputs;
5651 /* Compose SAMPLE action for sFlow. */
5653 compose_sflow_action(const struct ofproto_dpif *ofproto,
5654 struct ofpbuf *odp_actions,
5655 const struct flow *flow,
5658 uint32_t probability;
5659 union user_action_cookie cookie;
5660 size_t sample_offset, actions_offset;
5663 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5667 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5669 /* Number of packets out of UINT_MAX to sample. */
5670 probability = dpif_sflow_get_probability(ofproto->sflow);
5671 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5673 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5674 compose_sflow_cookie(ofproto, htons(0), odp_port,
5675 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5676 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
5678 nl_msg_end_nested(odp_actions, actions_offset);
5679 nl_msg_end_nested(odp_actions, sample_offset);
5680 return cookie_offset;
5683 /* SAMPLE action must be first action in any given list of actions.
5684 * At this point we do not have all information required to build it. So try to
5685 * build sample action as complete as possible. */
5687 add_sflow_action(struct action_xlate_ctx *ctx)
5689 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
5691 &ctx->flow, OVSP_NONE);
5692 ctx->sflow_odp_port = 0;
5693 ctx->sflow_n_outputs = 0;
5696 /* Fix SAMPLE action according to data collected while composing ODP actions.
5697 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
5698 * USERSPACE action's user-cookie which is required for sflow. */
5700 fix_sflow_action(struct action_xlate_ctx *ctx)
5702 const struct flow *base = &ctx->base_flow;
5703 union user_action_cookie *cookie;
5705 if (!ctx->user_cookie_offset) {
5709 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
5711 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
5713 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
5714 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
5718 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
5721 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
5722 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
5723 ovs_be64 flow_tun_id = ctx->flow.tunnel.tun_id;
5724 uint8_t flow_nw_tos = ctx->flow.nw_tos;
5725 struct priority_to_dscp *pdscp;
5726 uint32_t out_port, odp_port;
5728 /* If 'struct flow' gets additional metadata, we'll need to zero it out
5729 * before traversing a patch port. */
5730 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 18);
5733 xlate_report(ctx, "Nonexistent output port");
5735 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
5736 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
5738 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
5739 xlate_report(ctx, "STP not in forwarding state, skipping output");
5743 if (netdev_vport_is_patch(ofport->up.netdev)) {
5744 struct ofport_dpif *peer = ofport_get_peer(ofport);
5745 struct flow old_flow = ctx->flow;
5746 const struct ofproto_dpif *peer_ofproto;
5747 enum slow_path_reason special;
5748 struct ofport_dpif *in_port;
5751 xlate_report(ctx, "Nonexistent patch port peer");
5755 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5756 if (peer_ofproto->backer != ctx->ofproto->backer) {
5757 xlate_report(ctx, "Patch port peer on a different datapath");
5761 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
5762 ctx->flow.in_port = peer->up.ofp_port;
5763 ctx->flow.metadata = htonll(0);
5764 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
5765 memset(ctx->flow.regs, 0, sizeof ctx->flow.regs);
5767 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5768 special = process_special(ctx->ofproto, &ctx->flow, in_port,
5771 ctx->slow |= special;
5772 } else if (!in_port || may_receive(in_port, ctx)) {
5773 if (!in_port || stp_forward_in_state(in_port->stp_state)) {
5774 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
5776 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
5777 * learning action look at the packet, then drop it. */
5778 struct flow old_base_flow = ctx->base_flow;
5779 size_t old_size = ctx->odp_actions->size;
5780 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
5781 ctx->base_flow = old_base_flow;
5782 ctx->odp_actions->size = old_size;
5786 ctx->flow = old_flow;
5787 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5789 if (ctx->resubmit_stats) {
5790 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
5791 netdev_vport_inc_rx(peer->up.netdev, ctx->resubmit_stats);
5797 pdscp = get_priority(ofport, ctx->flow.skb_priority);
5799 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5800 ctx->flow.nw_tos |= pdscp->dscp;
5803 odp_port = ofp_port_to_odp_port(ctx->ofproto, ofp_port);
5804 if (ofport->tnl_port) {
5805 odp_port = tnl_port_send(ofport->tnl_port, &ctx->flow);
5806 if (odp_port == OVSP_NONE) {
5807 xlate_report(ctx, "Tunneling decided against output");
5811 if (ctx->resubmit_stats) {
5812 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
5814 out_port = odp_port;
5815 commit_odp_tunnel_action(&ctx->flow, &ctx->base_flow,
5818 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
5819 ctx->flow.vlan_tci);
5820 if (out_port != odp_port) {
5821 ctx->flow.vlan_tci = htons(0);
5823 ctx->flow.skb_mark &= ~IPSEC_MARK;
5825 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
5826 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
5828 ctx->sflow_odp_port = odp_port;
5829 ctx->sflow_n_outputs++;
5830 ctx->nf_output_iface = ofp_port;
5831 ctx->flow.tunnel.tun_id = flow_tun_id;
5832 ctx->flow.vlan_tci = flow_vlan_tci;
5833 ctx->flow.nw_tos = flow_nw_tos;
5837 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
5839 compose_output_action__(ctx, ofp_port, true);
5843 xlate_table_action(struct action_xlate_ctx *ctx,
5844 uint16_t in_port, uint8_t table_id, bool may_packet_in)
5846 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
5847 struct ofproto_dpif *ofproto = ctx->ofproto;
5848 struct rule_dpif *rule;
5849 uint16_t old_in_port;
5850 uint8_t old_table_id;
5852 old_table_id = ctx->table_id;
5853 ctx->table_id = table_id;
5855 /* Look up a flow with 'in_port' as the input port. */
5856 old_in_port = ctx->flow.in_port;
5857 ctx->flow.in_port = in_port;
5858 rule = rule_dpif_lookup__(ofproto, &ctx->flow, table_id);
5861 if (table_id > 0 && table_id < N_TABLES) {
5862 struct table_dpif *table = &ofproto->tables[table_id];
5863 if (table->other_table) {
5864 ctx->tags |= (rule && rule->tag
5866 : rule_calculate_tag(&ctx->flow,
5867 &table->other_table->mask,
5872 /* Restore the original input port. Otherwise OFPP_NORMAL and
5873 * OFPP_IN_PORT will have surprising behavior. */
5874 ctx->flow.in_port = old_in_port;
5876 if (ctx->resubmit_hook) {
5877 ctx->resubmit_hook(ctx, rule);
5880 if (rule == NULL && may_packet_in) {
5882 * check if table configuration flags
5883 * OFPTC_TABLE_MISS_CONTROLLER, default.
5884 * OFPTC_TABLE_MISS_CONTINUE,
5885 * OFPTC_TABLE_MISS_DROP
5886 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
5888 rule = rule_dpif_miss_rule(ofproto, &ctx->flow);
5892 struct rule_dpif *old_rule = ctx->rule;
5894 if (ctx->resubmit_stats) {
5895 rule_credit_stats(rule, ctx->resubmit_stats);
5900 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
5901 ctx->rule = old_rule;
5905 ctx->table_id = old_table_id;
5907 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5909 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
5910 MAX_RESUBMIT_RECURSION);
5911 ctx->max_resubmit_trigger = true;
5916 xlate_ofpact_resubmit(struct action_xlate_ctx *ctx,
5917 const struct ofpact_resubmit *resubmit)
5922 in_port = resubmit->in_port;
5923 if (in_port == OFPP_IN_PORT) {
5924 in_port = ctx->flow.in_port;
5927 table_id = resubmit->table_id;
5928 if (table_id == 255) {
5929 table_id = ctx->table_id;
5932 xlate_table_action(ctx, in_port, table_id, false);
5936 flood_packets(struct action_xlate_ctx *ctx, bool all)
5938 struct ofport_dpif *ofport;
5940 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
5941 uint16_t ofp_port = ofport->up.ofp_port;
5943 if (ofp_port == ctx->flow.in_port) {
5948 compose_output_action__(ctx, ofp_port, false);
5949 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
5950 compose_output_action(ctx, ofp_port);
5954 ctx->nf_output_iface = NF_OUT_FLOOD;
5958 execute_controller_action(struct action_xlate_ctx *ctx, int len,
5959 enum ofp_packet_in_reason reason,
5960 uint16_t controller_id)
5962 struct ofputil_packet_in pin;
5963 struct ofpbuf *packet;
5965 ctx->slow |= SLOW_CONTROLLER;
5970 packet = ofpbuf_clone(ctx->packet);
5972 if (packet->l2 && packet->l3) {
5973 struct eth_header *eh;
5975 eth_pop_vlan(packet);
5978 /* If the Ethernet type is less than ETH_TYPE_MIN, it's likely an 802.2
5979 * LLC frame. Calculating the Ethernet type of these frames is more
5980 * trouble than seems appropriate for a simple assertion. */
5981 ovs_assert(ntohs(eh->eth_type) < ETH_TYPE_MIN
5982 || eh->eth_type == ctx->flow.dl_type);
5984 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
5985 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
5987 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
5988 eth_push_vlan(packet, ctx->flow.vlan_tci);
5992 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5993 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
5994 ctx->flow.nw_tos, ctx->flow.nw_ttl);
5998 if (ctx->flow.nw_proto == IPPROTO_TCP) {
5999 packet_set_tcp_port(packet, ctx->flow.tp_src,
6001 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
6002 packet_set_udp_port(packet, ctx->flow.tp_src,
6009 pin.packet = packet->data;
6010 pin.packet_len = packet->size;
6011 pin.reason = reason;
6012 pin.controller_id = controller_id;
6013 pin.table_id = ctx->table_id;
6014 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
6017 flow_get_metadata(&ctx->flow, &pin.fmd);
6019 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
6020 ofpbuf_delete(packet);
6024 compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
6026 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
6027 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
6031 if (ctx->flow.nw_ttl > 1) {
6037 for (i = 0; i < ids->n_controllers; i++) {
6038 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
6042 /* Stop processing for current table. */
6048 xlate_output_action(struct action_xlate_ctx *ctx,
6049 uint16_t port, uint16_t max_len, bool may_packet_in)
6051 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
6053 ctx->nf_output_iface = NF_OUT_DROP;
6057 compose_output_action(ctx, ctx->flow.in_port);
6060 xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in);
6066 flood_packets(ctx, false);
6069 flood_packets(ctx, true);
6071 case OFPP_CONTROLLER:
6072 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
6078 if (port != ctx->flow.in_port) {
6079 compose_output_action(ctx, port);
6081 xlate_report(ctx, "skipping output to input port");
6086 if (prev_nf_output_iface == NF_OUT_FLOOD) {
6087 ctx->nf_output_iface = NF_OUT_FLOOD;
6088 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
6089 ctx->nf_output_iface = prev_nf_output_iface;
6090 } else if (prev_nf_output_iface != NF_OUT_DROP &&
6091 ctx->nf_output_iface != NF_OUT_FLOOD) {
6092 ctx->nf_output_iface = NF_OUT_MULTI;
6097 xlate_output_reg_action(struct action_xlate_ctx *ctx,
6098 const struct ofpact_output_reg *or)
6100 uint64_t port = mf_get_subfield(&or->src, &ctx->flow);
6101 if (port <= UINT16_MAX) {
6102 xlate_output_action(ctx, port, or->max_len, false);
6107 xlate_enqueue_action(struct action_xlate_ctx *ctx,
6108 const struct ofpact_enqueue *enqueue)
6110 uint16_t ofp_port = enqueue->port;
6111 uint32_t queue_id = enqueue->queue;
6112 uint32_t flow_priority, priority;
6115 /* Translate queue to priority. */
6116 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6117 queue_id, &priority);
6119 /* Fall back to ordinary output action. */
6120 xlate_output_action(ctx, enqueue->port, 0, false);
6124 /* Check output port. */
6125 if (ofp_port == OFPP_IN_PORT) {
6126 ofp_port = ctx->flow.in_port;
6127 } else if (ofp_port == ctx->flow.in_port) {
6131 /* Add datapath actions. */
6132 flow_priority = ctx->flow.skb_priority;
6133 ctx->flow.skb_priority = priority;
6134 compose_output_action(ctx, ofp_port);
6135 ctx->flow.skb_priority = flow_priority;
6137 /* Update NetFlow output port. */
6138 if (ctx->nf_output_iface == NF_OUT_DROP) {
6139 ctx->nf_output_iface = ofp_port;
6140 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
6141 ctx->nf_output_iface = NF_OUT_MULTI;
6146 xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id)
6148 uint32_t skb_priority;
6150 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6151 queue_id, &skb_priority)) {
6152 ctx->flow.skb_priority = skb_priority;
6154 /* Couldn't translate queue to a priority. Nothing to do. A warning
6155 * has already been logged. */
6159 struct xlate_reg_state {
6165 xlate_autopath(struct action_xlate_ctx *ctx,
6166 const struct ofpact_autopath *ap)
6168 uint16_t ofp_port = ap->port;
6169 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
6171 if (!port || !port->bundle) {
6172 ofp_port = OFPP_NONE;
6173 } else if (port->bundle->bond) {
6174 /* Autopath does not support VLAN hashing. */
6175 struct ofport_dpif *slave = bond_choose_output_slave(
6176 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
6178 ofp_port = slave->up.ofp_port;
6181 nxm_reg_load(&ap->dst, ofp_port, &ctx->flow);
6185 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
6187 struct ofproto_dpif *ofproto = ofproto_;
6188 struct ofport_dpif *port;
6198 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
6201 port = get_ofp_port(ofproto, ofp_port);
6202 return port ? port->may_enable : false;
6207 xlate_bundle_action(struct action_xlate_ctx *ctx,
6208 const struct ofpact_bundle *bundle)
6212 port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto);
6213 if (bundle->dst.field) {
6214 nxm_reg_load(&bundle->dst, port, &ctx->flow);
6216 xlate_output_action(ctx, port, 0, false);
6221 xlate_learn_action(struct action_xlate_ctx *ctx,
6222 const struct ofpact_learn *learn)
6224 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
6225 struct ofputil_flow_mod fm;
6226 uint64_t ofpacts_stub[1024 / 8];
6227 struct ofpbuf ofpacts;
6230 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
6231 learn_execute(learn, &ctx->flow, &fm, &ofpacts);
6233 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
6234 if (error && !VLOG_DROP_WARN(&rl)) {
6235 VLOG_WARN("learning action failed to modify flow table (%s)",
6236 ofperr_get_name(error));
6239 ofpbuf_uninit(&ofpacts);
6242 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
6243 * means "infinite". */
6245 reduce_timeout(uint16_t max, uint16_t *timeout)
6247 if (max && (!*timeout || *timeout > max)) {
6253 xlate_fin_timeout(struct action_xlate_ctx *ctx,
6254 const struct ofpact_fin_timeout *oft)
6256 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
6257 struct rule_dpif *rule = ctx->rule;
6259 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
6260 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
6265 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
6267 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
6268 ? OFPUTIL_PC_NO_RECV_STP
6269 : OFPUTIL_PC_NO_RECV)) {
6273 /* Only drop packets here if both forwarding and learning are
6274 * disabled. If just learning is enabled, we need to have
6275 * OFPP_NORMAL and the learning action have a look at the packet
6276 * before we can drop it. */
6277 if (!stp_forward_in_state(port->stp_state)
6278 && !stp_learn_in_state(port->stp_state)) {
6286 tunnel_ecn_ok(struct action_xlate_ctx *ctx)
6288 if (is_ip_any(&ctx->base_flow)
6289 && (ctx->base_flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
6290 if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
6291 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
6292 " but is not ECN capable");
6295 /* Set the ECN CE value in the tunneled packet. */
6296 ctx->flow.nw_tos |= IP_ECN_CE;
6304 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
6305 struct action_xlate_ctx *ctx)
6307 bool was_evictable = true;
6308 const struct ofpact *a;
6311 /* Don't let the rule we're working on get evicted underneath us. */
6312 was_evictable = ctx->rule->up.evictable;
6313 ctx->rule->up.evictable = false;
6315 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
6316 struct ofpact_controller *controller;
6317 const struct ofpact_metadata *metadata;
6325 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
6326 ofpact_get_OUTPUT(a)->max_len, true);
6329 case OFPACT_CONTROLLER:
6330 controller = ofpact_get_CONTROLLER(a);
6331 execute_controller_action(ctx, controller->max_len,
6333 controller->controller_id);
6336 case OFPACT_ENQUEUE:
6337 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
6340 case OFPACT_SET_VLAN_VID:
6341 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
6342 ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
6346 case OFPACT_SET_VLAN_PCP:
6347 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
6348 ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
6353 case OFPACT_STRIP_VLAN:
6354 ctx->flow.vlan_tci = htons(0);
6357 case OFPACT_PUSH_VLAN:
6358 /* XXX 802.1AD(QinQ) */
6359 ctx->flow.vlan_tci = htons(VLAN_CFI);
6362 case OFPACT_SET_ETH_SRC:
6363 memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
6367 case OFPACT_SET_ETH_DST:
6368 memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
6372 case OFPACT_SET_IPV4_SRC:
6373 ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
6376 case OFPACT_SET_IPV4_DST:
6377 ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
6380 case OFPACT_SET_IPV4_DSCP:
6381 /* OpenFlow 1.0 only supports IPv4. */
6382 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6383 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
6384 ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
6388 case OFPACT_SET_L4_SRC_PORT:
6389 ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
6392 case OFPACT_SET_L4_DST_PORT:
6393 ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
6396 case OFPACT_RESUBMIT:
6397 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
6400 case OFPACT_SET_TUNNEL:
6401 ctx->flow.tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
6404 case OFPACT_SET_QUEUE:
6405 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
6408 case OFPACT_POP_QUEUE:
6409 ctx->flow.skb_priority = ctx->orig_skb_priority;
6412 case OFPACT_REG_MOVE:
6413 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow);
6416 case OFPACT_REG_LOAD:
6417 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow);
6420 case OFPACT_DEC_TTL:
6421 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6427 /* Nothing to do. */
6430 case OFPACT_MULTIPATH:
6431 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow);
6434 case OFPACT_AUTOPATH:
6435 xlate_autopath(ctx, ofpact_get_AUTOPATH(a));
6439 ctx->ofproto->has_bundle_action = true;
6440 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6443 case OFPACT_OUTPUT_REG:
6444 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6448 ctx->has_learn = true;
6449 if (ctx->may_learn) {
6450 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6458 case OFPACT_FIN_TIMEOUT:
6459 ctx->has_fin_timeout = true;
6460 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6463 case OFPACT_CLEAR_ACTIONS:
6465 * Nothing to do because writa-actions is not supported for now.
6466 * When writa-actions is supported, clear-actions also must
6467 * be supported at the same time.
6471 case OFPACT_WRITE_METADATA:
6472 metadata = ofpact_get_WRITE_METADATA(a);
6473 ctx->flow.metadata &= ~metadata->mask;
6474 ctx->flow.metadata |= metadata->metadata & metadata->mask;
6477 case OFPACT_GOTO_TABLE: {
6478 /* XXX remove recursion */
6479 /* It is assumed that goto-table is last action */
6480 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6481 ovs_assert(ctx->table_id < ogt->table_id);
6482 xlate_table_action(ctx, ctx->flow.in_port, ogt->table_id, true);
6490 ctx->rule->up.evictable = was_evictable;
6495 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
6496 struct ofproto_dpif *ofproto, const struct flow *flow,
6497 const struct initial_vals *initial_vals,
6498 struct rule_dpif *rule,
6499 uint8_t tcp_flags, const struct ofpbuf *packet)
6501 ovs_be64 initial_tun_id = flow->tunnel.tun_id;
6503 /* Flow initialization rules:
6504 * - 'base_flow' must match the kernel's view of the packet at the
6505 * time that action processing starts. 'flow' represents any
6506 * transformations we wish to make through actions.
6507 * - By default 'base_flow' and 'flow' are the same since the input
6508 * packet matches the output before any actions are applied.
6509 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
6510 * of the received packet as seen by the kernel. If we later output
6511 * to another device without any modifications this will cause us to
6512 * insert a new tag since the original one was stripped off by the
6514 * - Tunnel 'flow' is largely cleared when transitioning between
6515 * the input and output stages since it does not make sense to output
6516 * a packet with the exact headers that it was received with (i.e.
6517 * the destination IP is us). The one exception is the tun_id, which
6518 * is preserved to allow use in later resubmit lookups and loads into
6520 * - Tunnel 'base_flow' is completely cleared since that is what the
6521 * kernel does. If we wish to maintain the original values an action
6522 * needs to be generated. */
6524 ctx->ofproto = ofproto;
6526 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
6527 ctx->base_flow = ctx->flow;
6528 ctx->base_flow.vlan_tci = initial_vals->vlan_tci;
6529 ctx->base_flow.tunnel.ip_tos = initial_vals->tunnel_ip_tos;
6530 ctx->flow.tunnel.tun_id = initial_tun_id;
6532 ctx->packet = packet;
6533 ctx->may_learn = packet != NULL;
6534 ctx->tcp_flags = tcp_flags;
6535 ctx->resubmit_hook = NULL;
6536 ctx->report_hook = NULL;
6537 ctx->resubmit_stats = NULL;
6540 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
6541 * into datapath actions in 'odp_actions', using 'ctx'. */
6543 xlate_actions(struct action_xlate_ctx *ctx,
6544 const struct ofpact *ofpacts, size_t ofpacts_len,
6545 struct ofpbuf *odp_actions)
6547 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
6548 * that in the future we always keep a copy of the original flow for
6549 * tracing purposes. */
6550 static bool hit_resubmit_limit;
6552 enum slow_path_reason special;
6553 struct ofport_dpif *in_port;
6555 COVERAGE_INC(ofproto_dpif_xlate);
6557 ofpbuf_clear(odp_actions);
6558 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
6560 ctx->odp_actions = odp_actions;
6563 ctx->has_learn = false;
6564 ctx->has_normal = false;
6565 ctx->has_fin_timeout = false;
6566 ctx->nf_output_iface = NF_OUT_DROP;
6569 ctx->max_resubmit_trigger = false;
6570 ctx->orig_skb_priority = ctx->flow.skb_priority;
6574 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
6575 /* Do this conditionally because the copy is expensive enough that it
6576 * shows up in profiles.
6578 * We keep orig_flow in 'ctx' only because I couldn't make GCC 4.4
6579 * believe that I wasn't using it without initializing it if I kept it
6580 * in a local variable. */
6581 ctx->orig_flow = ctx->flow;
6584 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
6585 switch (ctx->ofproto->up.frag_handling) {
6586 case OFPC_FRAG_NORMAL:
6587 /* We must pretend that transport ports are unavailable. */
6588 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
6589 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
6592 case OFPC_FRAG_DROP:
6595 case OFPC_FRAG_REASM:
6598 case OFPC_FRAG_NX_MATCH:
6599 /* Nothing to do. */
6602 case OFPC_INVALID_TTL_TO_CONTROLLER:
6607 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
6608 special = process_special(ctx->ofproto, &ctx->flow, in_port, ctx->packet);
6610 ctx->slow |= special;
6612 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6613 struct initial_vals initial_vals;
6614 uint32_t local_odp_port;
6616 initial_vals.vlan_tci = ctx->base_flow.vlan_tci;
6617 initial_vals.tunnel_ip_tos = ctx->base_flow.tunnel.ip_tos;
6619 add_sflow_action(ctx);
6621 if (tunnel_ecn_ok(ctx) && (!in_port || may_receive(in_port, ctx))) {
6622 do_xlate_actions(ofpacts, ofpacts_len, ctx);
6624 /* We've let OFPP_NORMAL and the learning action look at the
6625 * packet, so drop it now if forwarding is disabled. */
6626 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
6627 ofpbuf_clear(ctx->odp_actions);
6628 add_sflow_action(ctx);
6632 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
6633 if (!hit_resubmit_limit) {
6634 /* We didn't record the original flow. Make sure we do from
6636 hit_resubmit_limit = true;
6637 } else if (!VLOG_DROP_ERR(&trace_rl)) {
6638 struct ds ds = DS_EMPTY_INITIALIZER;
6640 ofproto_trace(ctx->ofproto, &ctx->orig_flow, ctx->packet,
6641 &initial_vals, &ds);
6642 VLOG_ERR("Trace triggered by excessive resubmit "
6643 "recursion:\n%s", ds_cstr(&ds));
6648 local_odp_port = ofp_port_to_odp_port(ctx->ofproto, OFPP_LOCAL);
6649 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
6651 ctx->odp_actions->data,
6652 ctx->odp_actions->size)) {
6653 ctx->slow |= SLOW_IN_BAND;
6655 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
6657 compose_output_action(ctx, OFPP_LOCAL);
6660 if (ctx->ofproto->has_mirrors) {
6661 add_mirror_actions(ctx, &ctx->orig_flow);
6663 fix_sflow_action(ctx);
6667 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
6668 * into datapath actions, using 'ctx', and discards the datapath actions. */
6670 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
6671 const struct ofpact *ofpacts,
6674 uint64_t odp_actions_stub[1024 / 8];
6675 struct ofpbuf odp_actions;
6677 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
6678 xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions);
6679 ofpbuf_uninit(&odp_actions);
6683 xlate_report(struct action_xlate_ctx *ctx, const char *s)
6685 if (ctx->report_hook) {
6686 ctx->report_hook(ctx, s);
6690 /* OFPP_NORMAL implementation. */
6692 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
6694 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
6695 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
6696 * the bundle on which the packet was received, returns the VLAN to which the
6699 * Both 'vid' and the return value are in the range 0...4095. */
6701 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
6703 switch (in_bundle->vlan_mode) {
6704 case PORT_VLAN_ACCESS:
6705 return in_bundle->vlan;
6708 case PORT_VLAN_TRUNK:
6711 case PORT_VLAN_NATIVE_UNTAGGED:
6712 case PORT_VLAN_NATIVE_TAGGED:
6713 return vid ? vid : in_bundle->vlan;
6720 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
6721 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
6724 * 'vid' should be the VID obtained from the 802.1Q header that was received as
6725 * part of a packet (specify 0 if there was no 802.1Q header), in the range
6728 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
6730 /* Allow any VID on the OFPP_NONE port. */
6731 if (in_bundle == &ofpp_none_bundle) {
6735 switch (in_bundle->vlan_mode) {
6736 case PORT_VLAN_ACCESS:
6739 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6740 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
6741 "packet received on port %s configured as VLAN "
6742 "%"PRIu16" access port",
6743 in_bundle->ofproto->up.name, vid,
6744 in_bundle->name, in_bundle->vlan);
6750 case PORT_VLAN_NATIVE_UNTAGGED:
6751 case PORT_VLAN_NATIVE_TAGGED:
6753 /* Port must always carry its native VLAN. */
6757 case PORT_VLAN_TRUNK:
6758 if (!ofbundle_includes_vlan(in_bundle, vid)) {
6760 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6761 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
6762 "received on port %s not configured for trunking "
6764 in_bundle->ofproto->up.name, vid,
6765 in_bundle->name, vid);
6777 /* Given 'vlan', the VLAN that a packet belongs to, and
6778 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
6779 * that should be included in the 802.1Q header. (If the return value is 0,
6780 * then the 802.1Q header should only be included in the packet if there is a
6783 * Both 'vlan' and the return value are in the range 0...4095. */
6785 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
6787 switch (out_bundle->vlan_mode) {
6788 case PORT_VLAN_ACCESS:
6791 case PORT_VLAN_TRUNK:
6792 case PORT_VLAN_NATIVE_TAGGED:
6795 case PORT_VLAN_NATIVE_UNTAGGED:
6796 return vlan == out_bundle->vlan ? 0 : vlan;
6804 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
6807 struct ofport_dpif *port;
6809 ovs_be16 tci, old_tci;
6811 vid = output_vlan_to_vid(out_bundle, vlan);
6812 if (!out_bundle->bond) {
6813 port = ofbundle_get_a_port(out_bundle);
6815 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
6818 /* No slaves enabled, so drop packet. */
6823 old_tci = ctx->flow.vlan_tci;
6825 if (tci || out_bundle->use_priority_tags) {
6826 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
6828 tci |= htons(VLAN_CFI);
6831 ctx->flow.vlan_tci = tci;
6833 compose_output_action(ctx, port->up.ofp_port);
6834 ctx->flow.vlan_tci = old_tci;
6838 mirror_mask_ffs(mirror_mask_t mask)
6840 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
6845 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
6847 return (bundle->vlan_mode != PORT_VLAN_ACCESS
6848 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
6852 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
6854 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
6857 /* Returns an arbitrary interface within 'bundle'. */
6858 static struct ofport_dpif *
6859 ofbundle_get_a_port(const struct ofbundle *bundle)
6861 return CONTAINER_OF(list_front(&bundle->ports),
6862 struct ofport_dpif, bundle_node);
6866 vlan_is_mirrored(const struct ofmirror *m, int vlan)
6868 return !m->vlans || bitmap_is_set(m->vlans, vlan);
6872 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
6874 struct ofproto_dpif *ofproto = ctx->ofproto;
6875 mirror_mask_t mirrors;
6876 struct ofbundle *in_bundle;
6879 const struct nlattr *a;
6882 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
6883 ctx->packet != NULL, NULL);
6887 mirrors = in_bundle->src_mirrors;
6889 /* Drop frames on bundles reserved for mirroring. */
6890 if (in_bundle->mirror_out) {
6891 if (ctx->packet != NULL) {
6892 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6893 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
6894 "%s, which is reserved exclusively for mirroring",
6895 ctx->ofproto->up.name, in_bundle->name);
6901 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
6902 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
6905 vlan = input_vid_to_vlan(in_bundle, vid);
6907 /* Look at the output ports to check for destination selections. */
6909 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
6910 ctx->odp_actions->size) {
6911 enum ovs_action_attr type = nl_attr_type(a);
6912 struct ofport_dpif *ofport;
6914 if (type != OVS_ACTION_ATTR_OUTPUT) {
6918 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
6919 if (ofport && ofport->bundle) {
6920 mirrors |= ofport->bundle->dst_mirrors;
6928 /* Restore the original packet before adding the mirror actions. */
6929 ctx->flow = *orig_flow;
6934 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6936 if (!vlan_is_mirrored(m, vlan)) {
6937 mirrors = zero_rightmost_1bit(mirrors);
6941 mirrors &= ~m->dup_mirrors;
6942 ctx->mirrors |= m->dup_mirrors;
6944 output_normal(ctx, m->out, vlan);
6945 } else if (vlan != m->out_vlan
6946 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
6947 struct ofbundle *bundle;
6949 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
6950 if (ofbundle_includes_vlan(bundle, m->out_vlan)
6951 && !bundle->mirror_out) {
6952 output_normal(ctx, bundle, m->out_vlan);
6960 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
6961 uint64_t packets, uint64_t bytes)
6967 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
6970 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6973 /* In normal circumstances 'm' will not be NULL. However,
6974 * if mirrors are reconfigured, we can temporarily get out
6975 * of sync in facet_revalidate(). We could "correct" the
6976 * mirror list before reaching here, but doing that would
6977 * not properly account the traffic stats we've currently
6978 * accumulated for previous mirror configuration. */
6982 m->packet_count += packets;
6983 m->byte_count += bytes;
6987 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
6988 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
6989 * indicate this; newer upstream kernels use gratuitous ARP requests. */
6991 is_gratuitous_arp(const struct flow *flow)
6993 return (flow->dl_type == htons(ETH_TYPE_ARP)
6994 && eth_addr_is_broadcast(flow->dl_dst)
6995 && (flow->nw_proto == ARP_OP_REPLY
6996 || (flow->nw_proto == ARP_OP_REQUEST
6997 && flow->nw_src == flow->nw_dst)));
7001 update_learning_table(struct ofproto_dpif *ofproto,
7002 const struct flow *flow, int vlan,
7003 struct ofbundle *in_bundle)
7005 struct mac_entry *mac;
7007 /* Don't learn the OFPP_NONE port. */
7008 if (in_bundle == &ofpp_none_bundle) {
7012 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
7016 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
7017 if (is_gratuitous_arp(flow)) {
7018 /* We don't want to learn from gratuitous ARP packets that are
7019 * reflected back over bond slaves so we lock the learning table. */
7020 if (!in_bundle->bond) {
7021 mac_entry_set_grat_arp_lock(mac);
7022 } else if (mac_entry_is_grat_arp_locked(mac)) {
7027 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
7028 /* The log messages here could actually be useful in debugging,
7029 * so keep the rate limit relatively high. */
7030 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
7031 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
7032 "on port %s in VLAN %d",
7033 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
7034 in_bundle->name, vlan);
7036 mac->port.p = in_bundle;
7037 tag_set_add(&ofproto->backer->revalidate_set,
7038 mac_learning_changed(ofproto->ml, mac));
7042 static struct ofbundle *
7043 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
7044 bool warn, struct ofport_dpif **in_ofportp)
7046 struct ofport_dpif *ofport;
7048 /* Find the port and bundle for the received packet. */
7049 ofport = get_ofp_port(ofproto, in_port);
7051 *in_ofportp = ofport;
7053 if (ofport && ofport->bundle) {
7054 return ofport->bundle;
7057 /* Special-case OFPP_NONE, which a controller may use as the ingress
7058 * port for traffic that it is sourcing. */
7059 if (in_port == OFPP_NONE) {
7060 return &ofpp_none_bundle;
7063 /* Odd. A few possible reasons here:
7065 * - We deleted a port but there are still a few packets queued up
7068 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
7069 * we don't know about.
7071 * - The ofproto client didn't configure the port as part of a bundle.
7072 * This is particularly likely to happen if a packet was received on the
7073 * port after it was created, but before the client had a chance to
7074 * configure its bundle.
7077 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7079 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
7080 "port %"PRIu16, ofproto->up.name, in_port);
7085 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
7086 * dropped. Returns true if they may be forwarded, false if they should be
7089 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
7090 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
7092 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
7093 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
7094 * checked by input_vid_is_valid().
7096 * May also add tags to '*tags', although the current implementation only does
7097 * so in one special case.
7100 is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port,
7103 struct ofproto_dpif *ofproto = ctx->ofproto;
7104 struct flow *flow = &ctx->flow;
7105 struct ofbundle *in_bundle = in_port->bundle;
7107 /* Drop frames for reserved multicast addresses
7108 * only if forward_bpdu option is absent. */
7109 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
7110 xlate_report(ctx, "packet has reserved destination MAC, dropping");
7114 if (in_bundle->bond) {
7115 struct mac_entry *mac;
7117 switch (bond_check_admissibility(in_bundle->bond, in_port,
7118 flow->dl_dst, &ctx->tags)) {
7123 xlate_report(ctx, "bonding refused admissibility, dropping");
7126 case BV_DROP_IF_MOVED:
7127 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
7128 if (mac && mac->port.p != in_bundle &&
7129 (!is_gratuitous_arp(flow)
7130 || mac_entry_is_grat_arp_locked(mac))) {
7131 xlate_report(ctx, "SLB bond thinks this packet looped back, "
7143 xlate_normal(struct action_xlate_ctx *ctx)
7145 struct ofport_dpif *in_port;
7146 struct ofbundle *in_bundle;
7147 struct mac_entry *mac;
7151 ctx->has_normal = true;
7153 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
7154 ctx->packet != NULL, &in_port);
7156 xlate_report(ctx, "no input bundle, dropping");
7160 /* Drop malformed frames. */
7161 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
7162 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
7163 if (ctx->packet != NULL) {
7164 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7165 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
7166 "VLAN tag received on port %s",
7167 ctx->ofproto->up.name, in_bundle->name);
7169 xlate_report(ctx, "partial VLAN tag, dropping");
7173 /* Drop frames on bundles reserved for mirroring. */
7174 if (in_bundle->mirror_out) {
7175 if (ctx->packet != NULL) {
7176 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7177 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7178 "%s, which is reserved exclusively for mirroring",
7179 ctx->ofproto->up.name, in_bundle->name);
7181 xlate_report(ctx, "input port is mirror output port, dropping");
7186 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
7187 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
7188 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
7191 vlan = input_vid_to_vlan(in_bundle, vid);
7193 /* Check other admissibility requirements. */
7194 if (in_port && !is_admissible(ctx, in_port, vlan)) {
7198 /* Learn source MAC. */
7199 if (ctx->may_learn) {
7200 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
7203 /* Determine output bundle. */
7204 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
7207 if (mac->port.p != in_bundle) {
7208 xlate_report(ctx, "forwarding to learned port");
7209 output_normal(ctx, mac->port.p, vlan);
7211 xlate_report(ctx, "learned port is input port, dropping");
7214 struct ofbundle *bundle;
7216 xlate_report(ctx, "no learned MAC for destination, flooding");
7217 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
7218 if (bundle != in_bundle
7219 && ofbundle_includes_vlan(bundle, vlan)
7220 && bundle->floodable
7221 && !bundle->mirror_out) {
7222 output_normal(ctx, bundle, vlan);
7225 ctx->nf_output_iface = NF_OUT_FLOOD;
7229 /* Optimized flow revalidation.
7231 * It's a difficult problem, in general, to tell which facets need to have
7232 * their actions recalculated whenever the OpenFlow flow table changes. We
7233 * don't try to solve that general problem: for most kinds of OpenFlow flow
7234 * table changes, we recalculate the actions for every facet. This is
7235 * relatively expensive, but it's good enough if the OpenFlow flow table
7236 * doesn't change very often.
7238 * However, we can expect one particular kind of OpenFlow flow table change to
7239 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
7240 * of CPU on revalidating every facet whenever MAC learning modifies the flow
7241 * table, we add a special case that applies to flow tables in which every rule
7242 * has the same form (that is, the same wildcards), except that the table is
7243 * also allowed to have a single "catch-all" flow that matches all packets. We
7244 * optimize this case by tagging all of the facets that resubmit into the table
7245 * and invalidating the same tag whenever a flow changes in that table. The
7246 * end result is that we revalidate just the facets that need it (and sometimes
7247 * a few more, but not all of the facets or even all of the facets that
7248 * resubmit to the table modified by MAC learning). */
7250 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
7251 * into an OpenFlow table with the given 'basis'. */
7253 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
7256 if (minimask_is_catchall(mask)) {
7259 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
7260 return tag_create_deterministic(hash);
7264 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
7265 * taggability of that table.
7267 * This function must be called after *each* change to a flow table. If you
7268 * skip calling it on some changes then the pointer comparisons at the end can
7269 * be invalid if you get unlucky. For example, if a flow removal causes a
7270 * cls_table to be destroyed and then a flow insertion causes a cls_table with
7271 * different wildcards to be created with the same address, then this function
7272 * will incorrectly skip revalidation. */
7274 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
7276 struct table_dpif *table = &ofproto->tables[table_id];
7277 const struct oftable *oftable = &ofproto->up.tables[table_id];
7278 struct cls_table *catchall, *other;
7279 struct cls_table *t;
7281 catchall = other = NULL;
7283 switch (hmap_count(&oftable->cls.tables)) {
7285 /* We could tag this OpenFlow table but it would make the logic a
7286 * little harder and it's a corner case that doesn't seem worth it
7292 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
7293 if (cls_table_is_catchall(t)) {
7295 } else if (!other) {
7298 /* Indicate that we can't tag this by setting both tables to
7299 * NULL. (We know that 'catchall' is already NULL.) */
7306 /* Can't tag this table. */
7310 if (table->catchall_table != catchall || table->other_table != other) {
7311 table->catchall_table = catchall;
7312 table->other_table = other;
7313 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7317 /* Given 'rule' that has changed in some way (either it is a rule being
7318 * inserted, a rule being deleted, or a rule whose actions are being
7319 * modified), marks facets for revalidation to ensure that packets will be
7320 * forwarded correctly according to the new state of the flow table.
7322 * This function must be called after *each* change to a flow table. See
7323 * the comment on table_update_taggable() for more information. */
7325 rule_invalidate(const struct rule_dpif *rule)
7327 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
7329 table_update_taggable(ofproto, rule->up.table_id);
7331 if (!ofproto->backer->need_revalidate) {
7332 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
7334 if (table->other_table && rule->tag) {
7335 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
7337 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7343 set_frag_handling(struct ofproto *ofproto_,
7344 enum ofp_config_flags frag_handling)
7346 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7347 if (frag_handling != OFPC_FRAG_REASM) {
7348 ofproto->backer->need_revalidate = REV_RECONFIGURE;
7356 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
7357 const struct flow *flow,
7358 const struct ofpact *ofpacts, size_t ofpacts_len)
7360 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7361 struct initial_vals initial_vals;
7362 struct odputil_keybuf keybuf;
7363 struct dpif_flow_stats stats;
7367 struct action_xlate_ctx ctx;
7368 uint64_t odp_actions_stub[1024 / 8];
7369 struct ofpbuf odp_actions;
7371 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
7372 odp_flow_key_from_flow(&key, flow,
7373 ofp_port_to_odp_port(ofproto, flow->in_port));
7375 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
7377 initial_vals.vlan_tci = flow->vlan_tci;
7378 initial_vals.tunnel_ip_tos = 0;
7379 action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals, NULL,
7380 packet_get_tcp_flags(packet, flow), packet);
7381 ctx.resubmit_stats = &stats;
7383 ofpbuf_use_stub(&odp_actions,
7384 odp_actions_stub, sizeof odp_actions_stub);
7385 xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions);
7386 dpif_execute(ofproto->backer->dpif, key.data, key.size,
7387 odp_actions.data, odp_actions.size, packet);
7388 ofpbuf_uninit(&odp_actions);
7396 set_netflow(struct ofproto *ofproto_,
7397 const struct netflow_options *netflow_options)
7399 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7401 if (netflow_options) {
7402 if (!ofproto->netflow) {
7403 ofproto->netflow = netflow_create();
7405 return netflow_set_options(ofproto->netflow, netflow_options);
7407 netflow_destroy(ofproto->netflow);
7408 ofproto->netflow = NULL;
7414 get_netflow_ids(const struct ofproto *ofproto_,
7415 uint8_t *engine_type, uint8_t *engine_id)
7417 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7419 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
7423 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
7425 if (!facet_is_controller_flow(facet) &&
7426 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
7427 struct subfacet *subfacet;
7428 struct ofexpired expired;
7430 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
7431 if (subfacet->path == SF_FAST_PATH) {
7432 struct dpif_flow_stats stats;
7434 subfacet_reinstall(subfacet, &stats);
7435 subfacet_update_stats(subfacet, &stats);
7439 expired.flow = facet->flow;
7440 expired.packet_count = facet->packet_count;
7441 expired.byte_count = facet->byte_count;
7442 expired.used = facet->used;
7443 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
7448 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
7450 struct facet *facet;
7452 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7453 send_active_timeout(ofproto, facet);
7457 static struct ofproto_dpif *
7458 ofproto_dpif_lookup(const char *name)
7460 struct ofproto_dpif *ofproto;
7462 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
7463 hash_string(name, 0), &all_ofproto_dpifs) {
7464 if (!strcmp(ofproto->up.name, name)) {
7472 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
7473 const char *argv[], void *aux OVS_UNUSED)
7475 struct ofproto_dpif *ofproto;
7478 ofproto = ofproto_dpif_lookup(argv[1]);
7480 unixctl_command_reply_error(conn, "no such bridge");
7483 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7485 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7486 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7490 unixctl_command_reply(conn, "table successfully flushed");
7494 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
7495 const char *argv[], void *aux OVS_UNUSED)
7497 struct ds ds = DS_EMPTY_INITIALIZER;
7498 const struct ofproto_dpif *ofproto;
7499 const struct mac_entry *e;
7501 ofproto = ofproto_dpif_lookup(argv[1]);
7503 unixctl_command_reply_error(conn, "no such bridge");
7507 ds_put_cstr(&ds, " port VLAN MAC Age\n");
7508 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
7509 struct ofbundle *bundle = e->port.p;
7510 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
7511 ofbundle_get_a_port(bundle)->odp_port,
7512 e->vlan, ETH_ADDR_ARGS(e->mac),
7513 mac_entry_age(ofproto->ml, e));
7515 unixctl_command_reply(conn, ds_cstr(&ds));
7520 struct action_xlate_ctx ctx;
7526 trace_format_rule(struct ds *result, uint8_t table_id, int level,
7527 const struct rule_dpif *rule)
7529 ds_put_char_multiple(result, '\t', level);
7531 ds_put_cstr(result, "No match\n");
7535 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
7536 table_id, ntohll(rule->up.flow_cookie));
7537 cls_rule_format(&rule->up.cr, result);
7538 ds_put_char(result, '\n');
7540 ds_put_char_multiple(result, '\t', level);
7541 ds_put_cstr(result, "OpenFlow ");
7542 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
7543 ds_put_char(result, '\n');
7547 trace_format_flow(struct ds *result, int level, const char *title,
7548 struct trace_ctx *trace)
7550 ds_put_char_multiple(result, '\t', level);
7551 ds_put_format(result, "%s: ", title);
7552 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
7553 ds_put_cstr(result, "unchanged");
7555 flow_format(result, &trace->ctx.flow);
7556 trace->flow = trace->ctx.flow;
7558 ds_put_char(result, '\n');
7562 trace_format_regs(struct ds *result, int level, const char *title,
7563 struct trace_ctx *trace)
7567 ds_put_char_multiple(result, '\t', level);
7568 ds_put_format(result, "%s:", title);
7569 for (i = 0; i < FLOW_N_REGS; i++) {
7570 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
7572 ds_put_char(result, '\n');
7576 trace_format_odp(struct ds *result, int level, const char *title,
7577 struct trace_ctx *trace)
7579 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
7581 ds_put_char_multiple(result, '\t', level);
7582 ds_put_format(result, "%s: ", title);
7583 format_odp_actions(result, odp_actions->data, odp_actions->size);
7584 ds_put_char(result, '\n');
7588 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
7590 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7591 struct ds *result = trace->result;
7593 ds_put_char(result, '\n');
7594 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
7595 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
7596 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
7597 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
7601 trace_report(struct action_xlate_ctx *ctx, const char *s)
7603 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7604 struct ds *result = trace->result;
7606 ds_put_char_multiple(result, '\t', ctx->recurse);
7607 ds_put_cstr(result, s);
7608 ds_put_char(result, '\n');
7612 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
7613 void *aux OVS_UNUSED)
7615 const char *dpname = argv[1];
7616 struct ofproto_dpif *ofproto;
7617 struct ofpbuf odp_key;
7618 struct ofpbuf *packet;
7619 struct initial_vals initial_vals;
7625 ofpbuf_init(&odp_key, 0);
7628 ofproto = ofproto_dpif_lookup(dpname);
7630 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
7634 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
7635 /* ofproto/trace dpname flow [-generate] */
7636 const char *flow_s = argv[2];
7637 const char *generate_s = argv[3];
7639 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
7640 * flow. We guess which type it is based on whether 'flow_s' contains
7641 * an '(', since a datapath flow always contains '(') but an
7642 * OpenFlow-like flow should not (in fact it's allowed but I believe
7643 * that's not documented anywhere).
7645 * An alternative would be to try to parse 'flow_s' both ways, but then
7646 * it would be tricky giving a sensible error message. After all, do
7647 * you just say "syntax error" or do you present both error messages?
7648 * Both choices seem lousy. */
7649 if (strchr(flow_s, '(')) {
7652 /* Convert string to datapath key. */
7653 ofpbuf_init(&odp_key, 0);
7654 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
7656 unixctl_command_reply_error(conn, "Bad flow syntax");
7660 /* The user might have specified the wrong ofproto but within the
7661 * same backer. That's OK, ofproto_receive() can find the right
7663 if (ofproto_receive(ofproto->backer, NULL, odp_key.data,
7664 odp_key.size, &flow, NULL, &ofproto, NULL,
7666 unixctl_command_reply_error(conn, "Invalid flow");
7669 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
7673 error_s = parse_ofp_exact_flow(&flow, argv[2]);
7675 unixctl_command_reply_error(conn, error_s);
7680 initial_vals.vlan_tci = flow.vlan_tci;
7681 initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos;
7684 /* Generate a packet, if requested. */
7686 packet = ofpbuf_new(0);
7687 flow_compose(packet, &flow);
7689 } else if (argc == 7) {
7690 /* ofproto/trace dpname priority tun_id in_port mark packet */
7691 const char *priority_s = argv[2];
7692 const char *tun_id_s = argv[3];
7693 const char *in_port_s = argv[4];
7694 const char *mark_s = argv[5];
7695 const char *packet_s = argv[6];
7696 uint32_t in_port = atoi(in_port_s);
7697 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
7698 uint32_t priority = atoi(priority_s);
7699 uint32_t mark = atoi(mark_s);
7702 msg = eth_from_hex(packet_s, &packet);
7704 unixctl_command_reply_error(conn, msg);
7708 ds_put_cstr(&result, "Packet: ");
7709 s = ofp_packet_to_string(packet->data, packet->size);
7710 ds_put_cstr(&result, s);
7713 flow_extract(packet, priority, mark, NULL, in_port, &flow);
7714 flow.tunnel.tun_id = tun_id;
7715 initial_vals.vlan_tci = flow.vlan_tci;
7716 initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos;
7718 unixctl_command_reply_error(conn, "Bad command syntax");
7722 ofproto_trace(ofproto, &flow, packet, &initial_vals, &result);
7723 unixctl_command_reply(conn, ds_cstr(&result));
7726 ds_destroy(&result);
7727 ofpbuf_delete(packet);
7728 ofpbuf_uninit(&odp_key);
7732 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
7733 const struct ofpbuf *packet,
7734 const struct initial_vals *initial_vals, struct ds *ds)
7736 struct rule_dpif *rule;
7738 ds_put_cstr(ds, "Flow: ");
7739 flow_format(ds, flow);
7740 ds_put_char(ds, '\n');
7742 rule = rule_dpif_lookup(ofproto, flow);
7744 trace_format_rule(ds, 0, 0, rule);
7745 if (rule == ofproto->miss_rule) {
7746 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
7747 } else if (rule == ofproto->no_packet_in_rule) {
7748 ds_put_cstr(ds, "\nNo match, packets dropped because "
7749 "OFPPC_NO_PACKET_IN is set on in_port.\n");
7753 uint64_t odp_actions_stub[1024 / 8];
7754 struct ofpbuf odp_actions;
7756 struct trace_ctx trace;
7759 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
7762 ofpbuf_use_stub(&odp_actions,
7763 odp_actions_stub, sizeof odp_actions_stub);
7764 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_vals,
7765 rule, tcp_flags, packet);
7766 trace.ctx.resubmit_hook = trace_resubmit;
7767 trace.ctx.report_hook = trace_report;
7768 xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len,
7771 ds_put_char(ds, '\n');
7772 trace_format_flow(ds, 0, "Final flow", &trace);
7773 ds_put_cstr(ds, "Datapath actions: ");
7774 format_odp_actions(ds, odp_actions.data, odp_actions.size);
7775 ofpbuf_uninit(&odp_actions);
7777 if (trace.ctx.slow) {
7778 enum slow_path_reason slow;
7780 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
7781 "slow path because it:");
7782 for (slow = trace.ctx.slow; slow; ) {
7783 enum slow_path_reason bit = rightmost_1bit(slow);
7787 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
7790 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
7793 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
7796 ds_put_cstr(ds, "\n\t- Needs in-band special case "
7799 ds_put_cstr(ds, "\n\t (The datapath actions are "
7800 "incomplete--for complete actions, "
7801 "please supply a packet.)");
7804 case SLOW_CONTROLLER:
7805 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
7806 "to the OpenFlow controller.");
7809 ds_put_cstr(ds, "\n\t- Needs more specific matching "
7810 "than the datapath supports.");
7817 if (slow & ~SLOW_MATCH) {
7818 ds_put_cstr(ds, "\nThe datapath actions above do not reflect "
7819 "the special slow-path processing.");
7826 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7827 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7830 unixctl_command_reply(conn, NULL);
7834 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7835 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7838 unixctl_command_reply(conn, NULL);
7841 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
7842 * 'reply' describing the results. */
7844 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
7846 struct facet *facet;
7850 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7851 if (!facet_check_consistency(facet)) {
7856 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
7860 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
7861 ofproto->up.name, errors);
7863 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
7868 ofproto_dpif_self_check(struct unixctl_conn *conn,
7869 int argc, const char *argv[], void *aux OVS_UNUSED)
7871 struct ds reply = DS_EMPTY_INITIALIZER;
7872 struct ofproto_dpif *ofproto;
7875 ofproto = ofproto_dpif_lookup(argv[1]);
7877 unixctl_command_reply_error(conn, "Unknown ofproto (use "
7878 "ofproto/list for help)");
7881 ofproto_dpif_self_check__(ofproto, &reply);
7883 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7884 ofproto_dpif_self_check__(ofproto, &reply);
7888 unixctl_command_reply(conn, ds_cstr(&reply));
7892 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
7893 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
7894 * to destroy 'ofproto_shash' and free the returned value. */
7895 static const struct shash_node **
7896 get_ofprotos(struct shash *ofproto_shash)
7898 const struct ofproto_dpif *ofproto;
7900 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7901 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
7902 shash_add_nocopy(ofproto_shash, name, ofproto);
7905 return shash_sort(ofproto_shash);
7909 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
7910 const char *argv[] OVS_UNUSED,
7911 void *aux OVS_UNUSED)
7913 struct ds ds = DS_EMPTY_INITIALIZER;
7914 struct shash ofproto_shash;
7915 const struct shash_node **sorted_ofprotos;
7918 shash_init(&ofproto_shash);
7919 sorted_ofprotos = get_ofprotos(&ofproto_shash);
7920 for (i = 0; i < shash_count(&ofproto_shash); i++) {
7921 const struct shash_node *node = sorted_ofprotos[i];
7922 ds_put_format(&ds, "%s\n", node->name);
7925 shash_destroy(&ofproto_shash);
7926 free(sorted_ofprotos);
7928 unixctl_command_reply(conn, ds_cstr(&ds));
7933 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
7935 struct dpif_dp_stats s;
7936 const struct shash_node **ports;
7939 dpif_get_dp_stats(ofproto->backer->dpif, &s);
7941 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
7942 dpif_name(ofproto->backer->dpif));
7943 /* xxx It would be better to show bridge-specific stats instead
7944 * xxx of dp ones. */
7946 "\tlookups: hit:%"PRIu64" missed:%"PRIu64" lost:%"PRIu64"\n",
7947 s.n_hit, s.n_missed, s.n_lost);
7948 ds_put_format(ds, "\tflows: %zu\n",
7949 hmap_count(&ofproto->subfacets));
7951 ports = shash_sort(&ofproto->up.port_by_name);
7952 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
7953 const struct shash_node *node = ports[i];
7954 struct ofport *ofport = node->data;
7955 const char *name = netdev_get_name(ofport->netdev);
7956 const char *type = netdev_get_type(ofport->netdev);
7959 ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port);
7961 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
7962 if (odp_port != OVSP_NONE) {
7963 ds_put_format(ds, "%"PRIu32":", odp_port);
7965 ds_put_cstr(ds, "none:");
7968 if (strcmp(type, "system")) {
7969 struct netdev *netdev;
7972 ds_put_format(ds, " (%s", type);
7974 error = netdev_open(name, type, &netdev);
7979 error = netdev_get_config(netdev, &config);
7981 const struct smap_node **nodes;
7984 nodes = smap_sort(&config);
7985 for (i = 0; i < smap_count(&config); i++) {
7986 const struct smap_node *node = nodes[i];
7987 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
7988 node->key, node->value);
7992 smap_destroy(&config);
7994 netdev_close(netdev);
7996 ds_put_char(ds, ')');
7998 ds_put_char(ds, '\n');
8004 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
8005 const char *argv[], void *aux OVS_UNUSED)
8007 struct ds ds = DS_EMPTY_INITIALIZER;
8008 const struct ofproto_dpif *ofproto;
8012 for (i = 1; i < argc; i++) {
8013 ofproto = ofproto_dpif_lookup(argv[i]);
8015 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
8016 "for help)", argv[i]);
8017 unixctl_command_reply_error(conn, ds_cstr(&ds));
8020 show_dp_format(ofproto, &ds);
8023 struct shash ofproto_shash;
8024 const struct shash_node **sorted_ofprotos;
8027 shash_init(&ofproto_shash);
8028 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8029 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8030 const struct shash_node *node = sorted_ofprotos[i];
8031 show_dp_format(node->data, &ds);
8034 shash_destroy(&ofproto_shash);
8035 free(sorted_ofprotos);
8038 unixctl_command_reply(conn, ds_cstr(&ds));
8043 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
8044 int argc OVS_UNUSED, const char *argv[],
8045 void *aux OVS_UNUSED)
8047 struct ds ds = DS_EMPTY_INITIALIZER;
8048 const struct ofproto_dpif *ofproto;
8049 struct subfacet *subfacet;
8051 ofproto = ofproto_dpif_lookup(argv[1]);
8053 unixctl_command_reply_error(conn, "no such bridge");
8057 update_stats(ofproto->backer);
8059 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
8060 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
8062 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
8063 subfacet->dp_packet_count, subfacet->dp_byte_count);
8064 if (subfacet->used) {
8065 ds_put_format(&ds, "%.3fs",
8066 (time_msec() - subfacet->used) / 1000.0);
8068 ds_put_format(&ds, "never");
8070 if (subfacet->facet->tcp_flags) {
8071 ds_put_cstr(&ds, ", flags:");
8072 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
8075 ds_put_cstr(&ds, ", actions:");
8076 if (subfacet->slow) {
8077 uint64_t slow_path_stub[128 / 8];
8078 const struct nlattr *actions;
8081 compose_slow_path(ofproto, &subfacet->facet->flow, subfacet->slow,
8082 slow_path_stub, sizeof slow_path_stub,
8083 &actions, &actions_len);
8084 format_odp_actions(&ds, actions, actions_len);
8086 format_odp_actions(&ds, subfacet->actions, subfacet->actions_len);
8088 ds_put_char(&ds, '\n');
8091 unixctl_command_reply(conn, ds_cstr(&ds));
8096 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
8097 int argc OVS_UNUSED, const char *argv[],
8098 void *aux OVS_UNUSED)
8100 struct ds ds = DS_EMPTY_INITIALIZER;
8101 struct ofproto_dpif *ofproto;
8103 ofproto = ofproto_dpif_lookup(argv[1]);
8105 unixctl_command_reply_error(conn, "no such bridge");
8109 flush(&ofproto->up);
8111 unixctl_command_reply(conn, ds_cstr(&ds));
8116 ofproto_dpif_unixctl_init(void)
8118 static bool registered;
8124 unixctl_command_register(
8126 "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}",
8127 2, 6, ofproto_unixctl_trace, NULL);
8128 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
8129 ofproto_unixctl_fdb_flush, NULL);
8130 unixctl_command_register("fdb/show", "bridge", 1, 1,
8131 ofproto_unixctl_fdb_show, NULL);
8132 unixctl_command_register("ofproto/clog", "", 0, 0,
8133 ofproto_dpif_clog, NULL);
8134 unixctl_command_register("ofproto/unclog", "", 0, 0,
8135 ofproto_dpif_unclog, NULL);
8136 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
8137 ofproto_dpif_self_check, NULL);
8138 unixctl_command_register("dpif/dump-dps", "", 0, 0,
8139 ofproto_unixctl_dpif_dump_dps, NULL);
8140 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
8141 ofproto_unixctl_dpif_show, NULL);
8142 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
8143 ofproto_unixctl_dpif_dump_flows, NULL);
8144 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
8145 ofproto_unixctl_dpif_del_flows, NULL);
8148 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
8150 * This is deprecated. It is only for compatibility with broken device drivers
8151 * in old versions of Linux that do not properly support VLANs when VLAN
8152 * devices are not used. When broken device drivers are no longer in
8153 * widespread use, we will delete these interfaces. */
8156 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
8158 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
8159 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
8161 if (realdev_ofp_port == ofport->realdev_ofp_port
8162 && vid == ofport->vlandev_vid) {
8166 ofproto->backer->need_revalidate = REV_RECONFIGURE;
8168 if (ofport->realdev_ofp_port) {
8171 if (realdev_ofp_port && ofport->bundle) {
8172 /* vlandevs are enslaved to their realdevs, so they are not allowed to
8173 * themselves be part of a bundle. */
8174 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
8177 ofport->realdev_ofp_port = realdev_ofp_port;
8178 ofport->vlandev_vid = vid;
8180 if (realdev_ofp_port) {
8181 vsp_add(ofport, realdev_ofp_port, vid);
8188 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
8190 return hash_2words(realdev_ofp_port, vid);
8193 /* Returns the ODP port number of the Linux VLAN device that corresponds to
8194 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
8195 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
8196 * it would return the port number of eth0.9.
8198 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
8199 * function just returns its 'realdev_odp_port' argument. */
8201 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
8202 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
8204 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
8205 uint16_t realdev_ofp_port;
8206 int vid = vlan_tci_to_vid(vlan_tci);
8207 const struct vlan_splinter *vsp;
8209 realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port);
8210 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
8211 hash_realdev_vid(realdev_ofp_port, vid),
8212 &ofproto->realdev_vid_map) {
8213 if (vsp->realdev_ofp_port == realdev_ofp_port
8214 && vsp->vid == vid) {
8215 return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port);
8219 return realdev_odp_port;
8222 static struct vlan_splinter *
8223 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
8225 struct vlan_splinter *vsp;
8227 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
8228 &ofproto->vlandev_map) {
8229 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
8237 /* Returns the OpenFlow port number of the "real" device underlying the Linux
8238 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
8239 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
8240 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
8241 * eth0 and store 9 in '*vid'.
8243 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
8244 * VLAN device. Unless VLAN splinters are enabled, this is what this function
8247 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
8248 uint16_t vlandev_ofp_port, int *vid)
8250 if (!hmap_is_empty(&ofproto->vlandev_map)) {
8251 const struct vlan_splinter *vsp;
8253 vsp = vlandev_find(ofproto, vlandev_ofp_port);
8258 return vsp->realdev_ofp_port;
8264 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
8265 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
8266 * 'flow->in_port' to the "real" device backing the VLAN device, sets
8267 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
8268 * always the case unless VLAN splinters are enabled), returns false without
8269 * making any changes. */
8271 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
8276 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
8281 /* Cause the flow to be processed as if it came in on the real device with
8282 * the VLAN device's VLAN ID. */
8283 flow->in_port = realdev;
8284 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
8289 vsp_remove(struct ofport_dpif *port)
8291 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8292 struct vlan_splinter *vsp;
8294 vsp = vlandev_find(ofproto, port->up.ofp_port);
8296 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
8297 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
8300 port->realdev_ofp_port = 0;
8302 VLOG_ERR("missing vlan device record");
8307 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
8309 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8311 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
8312 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
8313 == realdev_ofp_port)) {
8314 struct vlan_splinter *vsp;
8316 vsp = xmalloc(sizeof *vsp);
8317 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
8318 hash_int(port->up.ofp_port, 0));
8319 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
8320 hash_realdev_vid(realdev_ofp_port, vid));
8321 vsp->realdev_ofp_port = realdev_ofp_port;
8322 vsp->vlandev_ofp_port = port->up.ofp_port;
8325 port->realdev_ofp_port = realdev_ofp_port;
8327 VLOG_ERR("duplicate vlan device record");
8332 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
8334 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
8335 return ofport ? ofport->odp_port : OVSP_NONE;
8338 static struct ofport_dpif *
8339 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
8341 struct ofport_dpif *port;
8343 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
8344 hash_int(odp_port, 0),
8345 &backer->odp_to_ofport_map) {
8346 if (port->odp_port == odp_port) {
8355 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
8357 struct ofport_dpif *port;
8359 port = odp_port_to_ofport(ofproto->backer, odp_port);
8360 if (port && &ofproto->up == port->up.ofproto) {
8361 return port->up.ofp_port;
8367 const struct ofproto_class ofproto_dpif_class = {
8402 port_is_lacp_current,
8403 NULL, /* rule_choose_table */
8410 rule_modify_actions,
8419 get_cfm_remote_mpids,
8424 get_stp_port_status,
8431 is_mirror_output_bundle,
8432 forward_bpdu_changed,
8433 set_mac_table_config,