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-ipfix.h"
51 #include "ofproto-dpif-sflow.h"
52 #include "poll-loop.h"
57 #include "unaligned.h"
59 #include "vlan-bitmap.h"
62 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
64 COVERAGE_DEFINE(ofproto_dpif_expired);
65 COVERAGE_DEFINE(ofproto_dpif_xlate);
66 COVERAGE_DEFINE(facet_changed_rule);
67 COVERAGE_DEFINE(facet_revalidate);
68 COVERAGE_DEFINE(facet_unexpected);
69 COVERAGE_DEFINE(facet_suppress);
71 /* Maximum depth of flow table recursion (due to resubmit actions) in a
72 * flow translation. */
73 #define MAX_RESUBMIT_RECURSION 64
75 /* Number of implemented OpenFlow tables. */
76 enum { N_TABLES = 255 };
77 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
78 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
90 * - Do include packets and bytes from facets that have been deleted or
91 * whose own statistics have been folded into the rule.
93 * - Do include packets and bytes sent "by hand" that were accounted to
94 * the rule without any facet being involved (this is a rare corner
95 * case in rule_execute()).
97 * - Do not include packet or bytes that can be obtained from any facet's
98 * packet_count or byte_count member or that can be obtained from the
99 * datapath by, e.g., dpif_flow_get() for any subfacet.
101 uint64_t packet_count; /* Number of packets received. */
102 uint64_t byte_count; /* Number of bytes received. */
104 tag_type tag; /* Caches rule_calculate_tag() result. */
106 struct list facets; /* List of "struct facet"s. */
109 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
111 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
114 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
115 const struct flow *);
116 static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *,
119 static struct rule_dpif *rule_dpif_miss_rule(struct ofproto_dpif *ofproto,
120 const struct flow *flow);
122 static void rule_credit_stats(struct rule_dpif *,
123 const struct dpif_flow_stats *);
124 static void flow_push_stats(struct facet *, const struct dpif_flow_stats *);
125 static tag_type rule_calculate_tag(const struct flow *,
126 const struct minimask *, uint32_t basis);
127 static void rule_invalidate(const struct rule_dpif *);
129 #define MAX_MIRRORS 32
130 typedef uint32_t mirror_mask_t;
131 #define MIRROR_MASK_C(X) UINT32_C(X)
132 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
134 struct ofproto_dpif *ofproto; /* Owning ofproto. */
135 size_t idx; /* In ofproto's "mirrors" array. */
136 void *aux; /* Key supplied by ofproto's client. */
137 char *name; /* Identifier for log messages. */
139 /* Selection criteria. */
140 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
141 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
142 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
144 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
145 struct ofbundle *out; /* Output port or NULL. */
146 int out_vlan; /* Output VLAN or -1. */
147 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
150 int64_t packet_count; /* Number of packets sent. */
151 int64_t byte_count; /* Number of bytes sent. */
154 static void mirror_destroy(struct ofmirror *);
155 static void update_mirror_stats(struct ofproto_dpif *ofproto,
156 mirror_mask_t mirrors,
157 uint64_t packets, uint64_t bytes);
160 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
161 struct ofproto_dpif *ofproto; /* Owning ofproto. */
162 void *aux; /* Key supplied by ofproto's client. */
163 char *name; /* Identifier for log messages. */
166 struct list ports; /* Contains "struct ofport"s. */
167 enum port_vlan_mode vlan_mode; /* VLAN mode */
168 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
169 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
170 * NULL if all VLANs are trunked. */
171 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
172 struct bond *bond; /* Nonnull iff more than one port. */
173 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
176 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
178 /* Port mirroring info. */
179 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
180 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
181 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
184 static void bundle_remove(struct ofport *);
185 static void bundle_update(struct ofbundle *);
186 static void bundle_destroy(struct ofbundle *);
187 static void bundle_del_port(struct ofport_dpif *);
188 static void bundle_run(struct ofbundle *);
189 static void bundle_wait(struct ofbundle *);
190 static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *,
191 uint16_t in_port, bool warn,
192 struct ofport_dpif **in_ofportp);
194 /* A controller may use OFPP_NONE as the ingress port to indicate that
195 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
196 * when an input bundle is needed for validation (e.g., mirroring or
197 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
198 * any 'port' structs, so care must be taken when dealing with it. */
199 static struct ofbundle ofpp_none_bundle = {
201 .vlan_mode = PORT_VLAN_TRUNK
204 static void stp_run(struct ofproto_dpif *ofproto);
205 static void stp_wait(struct ofproto_dpif *ofproto);
206 static int set_stp_port(struct ofport *,
207 const struct ofproto_port_stp_settings *);
209 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
211 struct action_xlate_ctx {
212 /* action_xlate_ctx_init() initializes these members. */
215 struct ofproto_dpif *ofproto;
217 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
218 * this flow when actions change header fields. */
221 /* stack for the push and pop actions.
222 * Each stack element is of the type "union mf_subvalue". */
224 union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)];
226 /* The packet corresponding to 'flow', or a null pointer if we are
227 * revalidating without a packet to refer to. */
228 const struct ofpbuf *packet;
230 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
231 * actions update the flow table?
233 * We want to update these tables if we are actually processing a packet,
234 * or if we are accounting for packets that the datapath has processed, but
235 * not if we are just revalidating. */
238 /* The rule that we are currently translating, or NULL. */
239 struct rule_dpif *rule;
241 /* Union of the set of TCP flags seen so far in this flow. (Used only by
242 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
246 /* If nonnull, flow translation calls this function just before executing a
247 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
248 * when the recursion depth is exceeded.
250 * 'rule' is the rule being submitted into. It will be null if the
251 * resubmit or OFPP_TABLE action didn't find a matching rule.
253 * This is normally null so the client has to set it manually after
254 * calling action_xlate_ctx_init(). */
255 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule);
257 /* If nonnull, flow translation calls this function to report some
258 * significant decision, e.g. to explain why OFPP_NORMAL translation
259 * dropped a packet. */
260 void (*report_hook)(struct action_xlate_ctx *, const char *s);
262 /* If nonnull, flow translation credits the specified statistics to each
263 * rule reached through a resubmit or OFPP_TABLE action.
265 * This is normally null so the client has to set it manually after
266 * calling action_xlate_ctx_init(). */
267 const struct dpif_flow_stats *resubmit_stats;
269 /* xlate_actions() initializes and uses these members. The client might want
270 * to look at them after it returns. */
272 struct ofpbuf *odp_actions; /* Datapath actions. */
273 tag_type tags; /* Tags associated with actions. */
274 enum slow_path_reason slow; /* 0 if fast path may be used. */
275 bool has_learn; /* Actions include NXAST_LEARN? */
276 bool has_normal; /* Actions output to OFPP_NORMAL? */
277 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
278 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
279 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
281 /* xlate_actions() initializes and uses these members, but the client has no
282 * reason to look at them. */
284 int recurse; /* Recursion level, via xlate_table_action. */
285 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
286 struct flow base_flow; /* Flow at the last commit. */
287 uint32_t orig_skb_priority; /* Priority when packet arrived. */
288 uint8_t table_id; /* OpenFlow table ID where flow was found. */
289 uint32_t sflow_n_outputs; /* Number of output ports. */
290 uint32_t sflow_odp_port; /* Output port for composing sFlow action. */
291 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
292 bool exit; /* No further actions should be processed. */
295 /* Initial values of fields of the packet that may be changed during
296 * flow processing and needed later. */
297 struct initial_vals {
298 /* This is the value of vlan_tci in the packet as actually received from
299 * dpif. This is the same as the facet's flow.vlan_tci unless the packet
300 * was received via a VLAN splinter. In that case, this value is 0
301 * (because the packet as actually received from the dpif had no 802.1Q
302 * tag) but the facet's flow.vlan_tci is set to the VLAN that the splinter
305 * This member should be removed when the VLAN splinters feature is no
309 /* If received on a tunnel, the IP TOS value of the tunnel. */
310 uint8_t tunnel_ip_tos;
313 static void action_xlate_ctx_init(struct action_xlate_ctx *,
314 struct ofproto_dpif *, const struct flow *,
315 const struct initial_vals *initial_vals,
317 uint8_t tcp_flags, const struct ofpbuf *);
318 static void xlate_actions(struct action_xlate_ctx *,
319 const struct ofpact *ofpacts, size_t ofpacts_len,
320 struct ofpbuf *odp_actions);
321 static void xlate_actions_for_side_effects(struct action_xlate_ctx *,
322 const struct ofpact *ofpacts,
324 static void xlate_table_action(struct action_xlate_ctx *, uint16_t in_port,
325 uint8_t table_id, bool may_packet_in);
327 static size_t put_userspace_action(const struct ofproto_dpif *,
328 struct ofpbuf *odp_actions,
330 const union user_action_cookie *,
333 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
334 enum slow_path_reason,
335 uint64_t *stub, size_t stub_size,
336 const struct nlattr **actionsp,
337 size_t *actions_lenp);
339 static void xlate_report(struct action_xlate_ctx *ctx, const char *s);
341 /* A subfacet (see "struct subfacet" below) has three possible installation
344 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
345 * case just after the subfacet is created, just before the subfacet is
346 * destroyed, or if the datapath returns an error when we try to install a
349 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
351 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
352 * ofproto_dpif is installed in the datapath.
355 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
356 SF_FAST_PATH, /* Full actions are installed. */
357 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
360 static const char *subfacet_path_to_string(enum subfacet_path);
362 /* A dpif flow and actions associated with a facet.
364 * See also the large comment on struct facet. */
367 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
368 struct list list_node; /* In struct facet's 'facets' list. */
369 struct facet *facet; /* Owning facet. */
371 enum odp_key_fitness key_fitness;
375 long long int used; /* Time last used; time created if not used. */
376 long long int created; /* Time created. */
378 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
379 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
383 * These should be essentially identical for every subfacet in a facet, but
384 * may differ in trivial ways due to VLAN splinters. */
385 size_t actions_len; /* Number of bytes in actions[]. */
386 struct nlattr *actions; /* Datapath actions. */
388 enum slow_path_reason slow; /* 0 if fast path may be used. */
389 enum subfacet_path path; /* Installed in datapath? */
391 /* Initial values of the packet that may be needed later. */
392 struct initial_vals initial_vals;
394 /* Datapath port the packet arrived on. This is needed to remove
395 * flows for ports that are no longer part of the bridge. Since the
396 * flow definition only has the OpenFlow port number and the port is
397 * no longer part of the bridge, we can't determine the datapath port
398 * number needed to delete the flow from the datapath. */
399 uint32_t odp_in_port;
402 #define SUBFACET_DESTROY_MAX_BATCH 50
404 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
406 static struct subfacet *subfacet_find(struct ofproto_dpif *,
407 const struct nlattr *key, size_t key_len,
409 static void subfacet_destroy(struct subfacet *);
410 static void subfacet_destroy__(struct subfacet *);
411 static void subfacet_destroy_batch(struct ofproto_dpif *,
412 struct subfacet **, int n);
413 static void subfacet_reset_dp_stats(struct subfacet *,
414 struct dpif_flow_stats *);
415 static void subfacet_update_time(struct subfacet *, long long int used);
416 static void subfacet_update_stats(struct subfacet *,
417 const struct dpif_flow_stats *);
418 static void subfacet_make_actions(struct subfacet *,
419 const struct ofpbuf *packet,
420 struct ofpbuf *odp_actions);
421 static int subfacet_install(struct subfacet *,
422 const struct nlattr *actions, size_t actions_len,
423 struct dpif_flow_stats *, enum slow_path_reason);
424 static void subfacet_uninstall(struct subfacet *);
426 static enum subfacet_path subfacet_want_path(enum slow_path_reason);
428 /* An exact-match instantiation of an OpenFlow flow.
430 * A facet associates a "struct flow", which represents the Open vSwitch
431 * userspace idea of an exact-match flow, with one or more subfacets. Each
432 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
433 * the facet. When the kernel module (or other dpif implementation) and Open
434 * vSwitch userspace agree on the definition of a flow key, there is exactly
435 * one subfacet per facet. If the dpif implementation supports more-specific
436 * flow matching than userspace, however, a facet can have more than one
437 * subfacet, each of which corresponds to some distinction in flow that
438 * userspace simply doesn't understand.
440 * Flow expiration works in terms of subfacets, so a facet must have at least
441 * one subfacet or it will never expire, leaking memory. */
444 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
445 struct list list_node; /* In owning rule's 'facets' list. */
446 struct rule_dpif *rule; /* Owning rule. */
449 struct list subfacets;
450 long long int used; /* Time last used; time created if not used. */
457 * - Do include packets and bytes sent "by hand", e.g. with
460 * - Do include packets and bytes that were obtained from the datapath
461 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
462 * DPIF_FP_ZERO_STATS).
464 * - Do not include packets or bytes that can be obtained from the
465 * datapath for any existing subfacet.
467 uint64_t packet_count; /* Number of packets received. */
468 uint64_t byte_count; /* Number of bytes received. */
470 /* Resubmit statistics. */
471 uint64_t prev_packet_count; /* Number of packets from last stats push. */
472 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
473 long long int prev_used; /* Used time from last stats push. */
476 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
477 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
478 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
480 /* Properties of datapath actions.
482 * Every subfacet has its own actions because actions can differ slightly
483 * between splintered and non-splintered subfacets due to the VLAN tag
484 * being initially different (present vs. absent). All of them have these
485 * properties in common so we just store one copy of them here. */
486 bool has_learn; /* Actions include NXAST_LEARN? */
487 bool has_normal; /* Actions output to OFPP_NORMAL? */
488 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
489 tag_type tags; /* Tags that would require revalidation. */
490 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
492 /* Storage for a single subfacet, to reduce malloc() time and space
493 * overhead. (A facet always has at least one subfacet and in the common
494 * case has exactly one subfacet. However, 'one_subfacet' may not
495 * always be valid, since it could have been removed after newer
496 * subfacets were pushed onto the 'subfacets' list.) */
497 struct subfacet one_subfacet;
499 long long int learn_rl; /* Rate limiter for facet_learn(). */
502 static struct facet *facet_create(struct rule_dpif *,
503 const struct flow *, uint32_t hash);
504 static void facet_remove(struct facet *);
505 static void facet_free(struct facet *);
507 static struct facet *facet_find(struct ofproto_dpif *,
508 const struct flow *, uint32_t hash);
509 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
510 const struct flow *, uint32_t hash);
511 static void facet_revalidate(struct facet *);
512 static bool facet_check_consistency(struct facet *);
514 static void facet_flush_stats(struct facet *);
516 static void facet_update_time(struct facet *, long long int used);
517 static void facet_reset_counters(struct facet *);
518 static void facet_push_stats(struct facet *);
519 static void facet_learn(struct facet *);
520 static void facet_account(struct facet *);
521 static void push_all_stats(void);
523 static struct subfacet *facet_get_subfacet(struct facet *);
525 static bool facet_is_controller_flow(struct facet *);
528 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
532 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
533 struct list bundle_node; /* In struct ofbundle's "ports" list. */
534 struct cfm *cfm; /* Connectivity Fault Management, if any. */
535 struct bfd *bfd; /* BFD, if any. */
536 tag_type tag; /* Tag associated with this port. */
537 bool may_enable; /* May be enabled in bonds. */
538 long long int carrier_seq; /* Carrier status changes. */
539 struct tnl_port *tnl_port; /* Tunnel handle, or null. */
542 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
543 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
544 long long int stp_state_entered;
546 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
548 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
550 * This is deprecated. It is only for compatibility with broken device
551 * drivers in old versions of Linux that do not properly support VLANs when
552 * VLAN devices are not used. When broken device drivers are no longer in
553 * widespread use, we will delete these interfaces. */
554 uint16_t realdev_ofp_port;
558 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
559 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
560 * traffic egressing the 'ofport' with that priority should be marked with. */
561 struct priority_to_dscp {
562 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
563 uint32_t priority; /* Priority of this queue (see struct flow). */
565 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
568 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
570 * This is deprecated. It is only for compatibility with broken device drivers
571 * in old versions of Linux that do not properly support VLANs when VLAN
572 * devices are not used. When broken device drivers are no longer in
573 * widespread use, we will delete these interfaces. */
574 struct vlan_splinter {
575 struct hmap_node realdev_vid_node;
576 struct hmap_node vlandev_node;
577 uint16_t realdev_ofp_port;
578 uint16_t vlandev_ofp_port;
582 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
583 uint32_t realdev, ovs_be16 vlan_tci);
584 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
585 static void vsp_remove(struct ofport_dpif *);
586 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
588 static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *,
590 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
593 static struct ofport_dpif *
594 ofport_dpif_cast(const struct ofport *ofport)
596 ovs_assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
597 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
600 static void port_run(struct ofport_dpif *);
601 static void port_run_fast(struct ofport_dpif *);
602 static void port_wait(struct ofport_dpif *);
603 static int set_cfm(struct ofport *, const struct cfm_settings *);
604 static void ofport_clear_priorities(struct ofport_dpif *);
605 static void run_fast_rl(void);
607 struct dpif_completion {
608 struct list list_node;
609 struct ofoperation *op;
612 /* Extra information about a classifier table.
613 * Currently used just for optimized flow revalidation. */
615 /* If either of these is nonnull, then this table has a form that allows
616 * flows to be tagged to avoid revalidating most flows for the most common
617 * kinds of flow table changes. */
618 struct cls_table *catchall_table; /* Table that wildcards all fields. */
619 struct cls_table *other_table; /* Table with any other wildcard set. */
620 uint32_t basis; /* Keeps each table's tags separate. */
623 /* Reasons that we might need to revalidate every facet, and corresponding
626 * A value of 0 means that there is no need to revalidate.
628 * It would be nice to have some cleaner way to integrate with coverage
629 * counters, but with only a few reasons I guess this is good enough for
631 enum revalidate_reason {
632 REV_RECONFIGURE = 1, /* Switch configuration changed. */
633 REV_STP, /* Spanning tree protocol port status change. */
634 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
635 REV_FLOW_TABLE, /* Flow table changed. */
636 REV_INCONSISTENCY /* Facet self-check failed. */
638 COVERAGE_DEFINE(rev_reconfigure);
639 COVERAGE_DEFINE(rev_stp);
640 COVERAGE_DEFINE(rev_port_toggled);
641 COVERAGE_DEFINE(rev_flow_table);
642 COVERAGE_DEFINE(rev_inconsistency);
644 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
645 * These are datapath flows which have no associated ofproto, if they did we
646 * would use facets. */
648 struct hmap_node hmap_node;
653 /* All datapaths of a given type share a single dpif backer instance. */
658 struct timer next_expiration;
659 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
661 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
663 /* Facet revalidation flags applying to facets which use this backer. */
664 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
665 struct tag_set revalidate_set; /* Revalidate only matching facets. */
667 struct hmap drop_keys; /* Set of dropped odp keys. */
670 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
671 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
673 static void drop_key_clear(struct dpif_backer *);
674 static struct ofport_dpif *
675 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
677 static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto,
679 struct avg_subfacet_rates {
680 double add_rate; /* Moving average of new flows created per minute. */
681 double del_rate; /* Moving average of flows deleted per minute. */
683 static void show_dp_rates(struct ds *ds, const char *heading,
684 const struct avg_subfacet_rates *rates);
685 static void exp_mavg(double *avg, int base, double new);
687 struct ofproto_dpif {
688 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
690 struct dpif_backer *backer;
692 /* Special OpenFlow rules. */
693 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
694 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
700 struct netflow *netflow;
701 struct dpif_sflow *sflow;
702 struct dpif_ipfix *ipfix;
703 struct hmap bundles; /* Contains "struct ofbundle"s. */
704 struct mac_learning *ml;
705 struct ofmirror *mirrors[MAX_MIRRORS];
707 bool has_bonded_bundles;
711 struct hmap subfacets;
712 struct governor *governor;
713 long long int consistency_rl;
716 struct table_dpif tables[N_TABLES];
718 /* Support for debugging async flow mods. */
719 struct list completions;
721 bool has_bundle_action; /* True when the first bundle action appears. */
722 struct netdev_stats stats; /* To account packets generated and consumed in
727 long long int stp_last_tick;
729 /* VLAN splinters. */
730 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
731 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
734 struct sset ports; /* Set of standard port names. */
735 struct sset ghost_ports; /* Ports with no datapath port. */
736 struct sset port_poll_set; /* Queued names for port_poll() reply. */
737 int port_poll_errno; /* Last errno for port_poll() reply. */
739 /* Per ofproto's dpif stats. */
743 /* Subfacet statistics.
745 * These keep track of the total number of subfacets added and deleted and
746 * flow life span. They are useful for computing the flow rates stats
747 * exposed via "ovs-appctl dpif/show". The goal is to learn about
748 * traffic patterns in ways that we can use later to improve Open vSwitch
749 * performance in new situations. */
750 long long int created; /* Time when it is created. */
751 unsigned int max_n_subfacet; /* Maximum number of flows */
753 /* The average number of subfacets... */
754 struct avg_subfacet_rates hourly; /* ...over the last hour. */
755 struct avg_subfacet_rates daily; /* ...over the last day. */
756 long long int last_minute; /* Last time 'hourly' was updated. */
758 /* Number of subfacets added or deleted since 'last_minute'. */
759 unsigned int subfacet_add_count;
760 unsigned int subfacet_del_count;
762 /* Number of subfacets added or deleted from 'created' to 'last_minute.' */
763 unsigned long long int total_subfacet_add_count;
764 unsigned long long int total_subfacet_del_count;
766 /* Sum of the number of milliseconds that each subfacet existed,
767 * over the subfacets that have been added and then later deleted. */
768 unsigned long long int total_subfacet_life_span;
770 /* Incremented by the number of currently existing subfacets, each
771 * time we pull statistics from the kernel. */
772 unsigned long long int total_subfacet_count;
774 /* Number of times we pull statistics from the kernel. */
775 unsigned long long int n_update_stats;
777 static unsigned long long int avg_subfacet_life_span(
778 const struct ofproto_dpif *);
779 static double avg_subfacet_count(const struct ofproto_dpif *ofproto);
780 static void update_moving_averages(struct ofproto_dpif *ofproto);
781 static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto,
783 static void update_max_subfacet_count(struct ofproto_dpif *ofproto);
785 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
786 * for debugging the asynchronous flow_mod implementation.) */
789 /* All existing ofproto_dpif instances, indexed by ->up.name. */
790 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
792 static void ofproto_dpif_unixctl_init(void);
794 static struct ofproto_dpif *
795 ofproto_dpif_cast(const struct ofproto *ofproto)
797 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
798 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
801 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
803 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
805 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
806 const struct ofpbuf *,
807 const struct initial_vals *, struct ds *);
809 /* Packet processing. */
810 static void update_learning_table(struct ofproto_dpif *,
811 const struct flow *, int vlan,
814 #define FLOW_MISS_MAX_BATCH 50
815 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
817 /* Flow expiration. */
818 static int expire(struct dpif_backer *);
821 static void send_netflow_active_timeouts(struct ofproto_dpif *);
824 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
825 static size_t compose_sflow_action(const struct ofproto_dpif *,
826 struct ofpbuf *odp_actions,
827 const struct flow *, uint32_t odp_port);
828 static void compose_ipfix_action(const struct ofproto_dpif *,
829 struct ofpbuf *odp_actions,
830 const struct flow *);
831 static void add_mirror_actions(struct action_xlate_ctx *ctx,
832 const struct flow *flow);
833 /* Global variables. */
834 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
836 /* Initial mappings of port to bridge mappings. */
837 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
839 /* Factory functions. */
842 init(const struct shash *iface_hints)
844 struct shash_node *node;
846 /* Make a local copy, since we don't own 'iface_hints' elements. */
847 SHASH_FOR_EACH(node, iface_hints) {
848 const struct iface_hint *orig_hint = node->data;
849 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
851 new_hint->br_name = xstrdup(orig_hint->br_name);
852 new_hint->br_type = xstrdup(orig_hint->br_type);
853 new_hint->ofp_port = orig_hint->ofp_port;
855 shash_add(&init_ofp_ports, node->name, new_hint);
860 enumerate_types(struct sset *types)
862 dp_enumerate_types(types);
866 enumerate_names(const char *type, struct sset *names)
868 struct ofproto_dpif *ofproto;
871 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
872 if (strcmp(type, ofproto->up.type)) {
875 sset_add(names, ofproto->up.name);
882 del(const char *type, const char *name)
887 error = dpif_open(name, type, &dpif);
889 error = dpif_delete(dpif);
896 port_open_type(const char *datapath_type, const char *port_type)
898 return dpif_port_open_type(datapath_type, port_type);
901 /* Type functions. */
903 static struct ofproto_dpif *
904 lookup_ofproto_dpif_by_port_name(const char *name)
906 struct ofproto_dpif *ofproto;
908 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
909 if (sset_contains(&ofproto->ports, name)) {
918 type_run(const char *type)
920 static long long int push_timer = LLONG_MIN;
921 struct dpif_backer *backer;
925 backer = shash_find_data(&all_dpif_backers, type);
927 /* This is not necessarily a problem, since backers are only
928 * created on demand. */
932 dpif_run(backer->dpif);
934 /* The most natural place to push facet statistics is when they're pulled
935 * from the datapath. However, when there are many flows in the datapath,
936 * this expensive operation can occur so frequently, that it reduces our
937 * ability to quickly set up flows. To reduce the cost, we push statistics
939 if (time_msec() > push_timer) {
940 push_timer = time_msec() + 2000;
944 if (backer->need_revalidate
945 || !tag_set_is_empty(&backer->revalidate_set)) {
946 struct tag_set revalidate_set = backer->revalidate_set;
947 bool need_revalidate = backer->need_revalidate;
948 struct ofproto_dpif *ofproto;
949 struct simap_node *node;
950 struct simap tmp_backers;
952 /* Handle tunnel garbage collection. */
953 simap_init(&tmp_backers);
954 simap_swap(&backer->tnl_backers, &tmp_backers);
956 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
957 struct ofport_dpif *iter;
959 if (backer != ofproto->backer) {
963 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
966 if (!iter->tnl_port) {
970 dp_port = netdev_vport_get_dpif_port(iter->up.netdev);
971 node = simap_find(&tmp_backers, dp_port);
973 simap_put(&backer->tnl_backers, dp_port, node->data);
974 simap_delete(&tmp_backers, node);
975 node = simap_find(&backer->tnl_backers, dp_port);
977 node = simap_find(&backer->tnl_backers, dp_port);
979 uint32_t odp_port = UINT32_MAX;
981 if (!dpif_port_add(backer->dpif, iter->up.netdev,
983 simap_put(&backer->tnl_backers, dp_port, odp_port);
984 node = simap_find(&backer->tnl_backers, dp_port);
989 iter->odp_port = node ? node->data : OVSP_NONE;
990 if (tnl_port_reconfigure(&iter->up, iter->odp_port,
992 backer->need_revalidate = REV_RECONFIGURE;
997 SIMAP_FOR_EACH (node, &tmp_backers) {
998 dpif_port_del(backer->dpif, node->data);
1000 simap_destroy(&tmp_backers);
1002 switch (backer->need_revalidate) {
1003 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
1004 case REV_STP: COVERAGE_INC(rev_stp); break;
1005 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
1006 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
1007 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
1010 if (backer->need_revalidate) {
1011 /* Clear the drop_keys in case we should now be accepting some
1012 * formerly dropped flows. */
1013 drop_key_clear(backer);
1016 /* Clear the revalidation flags. */
1017 tag_set_init(&backer->revalidate_set);
1018 backer->need_revalidate = 0;
1020 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1021 struct facet *facet, *next;
1023 if (ofproto->backer != backer) {
1027 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
1029 || tag_set_intersects(&revalidate_set, facet->tags)) {
1030 facet_revalidate(facet);
1037 if (timer_expired(&backer->next_expiration)) {
1038 int delay = expire(backer);
1039 timer_set_duration(&backer->next_expiration, delay);
1042 /* Check for port changes in the dpif. */
1043 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
1044 struct ofproto_dpif *ofproto;
1045 struct dpif_port port;
1047 /* Don't report on the datapath's device. */
1048 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
1052 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1053 &all_ofproto_dpifs) {
1054 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
1059 ofproto = lookup_ofproto_dpif_by_port_name(devname);
1060 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
1061 /* The port was removed. If we know the datapath,
1062 * report it through poll_set(). If we don't, it may be
1063 * notifying us of a removal we initiated, so ignore it.
1064 * If there's a pending ENOBUFS, let it stand, since
1065 * everything will be reevaluated. */
1066 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
1067 sset_add(&ofproto->port_poll_set, devname);
1068 ofproto->port_poll_errno = 0;
1070 } else if (!ofproto) {
1071 /* The port was added, but we don't know with which
1072 * ofproto we should associate it. Delete it. */
1073 dpif_port_del(backer->dpif, port.port_no);
1075 dpif_port_destroy(&port);
1081 if (error != EAGAIN) {
1082 struct ofproto_dpif *ofproto;
1084 /* There was some sort of error, so propagate it to all
1085 * ofprotos that use this backer. */
1086 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1087 &all_ofproto_dpifs) {
1088 if (ofproto->backer == backer) {
1089 sset_clear(&ofproto->port_poll_set);
1090 ofproto->port_poll_errno = error;
1099 dpif_backer_run_fast(struct dpif_backer *backer, int max_batch)
1103 /* Handle one or more batches of upcalls, until there's nothing left to do
1104 * or until we do a fixed total amount of work.
1106 * We do work in batches because it can be much cheaper to set up a number
1107 * of flows and fire off their patches all at once. We do multiple batches
1108 * because in some cases handling a packet can cause another packet to be
1109 * queued almost immediately as part of the return flow. Both
1110 * optimizations can make major improvements on some benchmarks and
1111 * presumably for real traffic as well. */
1113 while (work < max_batch) {
1114 int retval = handle_upcalls(backer, max_batch - work);
1125 type_run_fast(const char *type)
1127 struct dpif_backer *backer;
1129 backer = shash_find_data(&all_dpif_backers, type);
1131 /* This is not necessarily a problem, since backers are only
1132 * created on demand. */
1136 return dpif_backer_run_fast(backer, FLOW_MISS_MAX_BATCH);
1142 static long long int port_rl = LLONG_MIN;
1143 static unsigned int backer_rl = 0;
1145 if (time_msec() >= port_rl) {
1146 struct ofproto_dpif *ofproto;
1147 struct ofport_dpif *ofport;
1149 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
1151 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1152 port_run_fast(ofport);
1155 port_rl = time_msec() + 200;
1158 /* XXX: We have to be careful not to do too much work in this function. If
1159 * we call dpif_backer_run_fast() too often, or with too large a batch,
1160 * performance improves signifcantly, but at a cost. It's possible for the
1161 * number of flows in the datapath to increase without bound, and for poll
1162 * loops to take 10s of seconds. The correct solution to this problem,
1163 * long term, is to separate flow miss handling into it's own thread so it
1164 * isn't affected by revalidations, and expirations. Until then, this is
1165 * the best we can do. */
1166 if (++backer_rl >= 10) {
1167 struct shash_node *node;
1170 SHASH_FOR_EACH (node, &all_dpif_backers) {
1171 dpif_backer_run_fast(node->data, 1);
1177 type_wait(const char *type)
1179 struct dpif_backer *backer;
1181 backer = shash_find_data(&all_dpif_backers, type);
1183 /* This is not necessarily a problem, since backers are only
1184 * created on demand. */
1188 timer_wait(&backer->next_expiration);
1191 /* Basic life-cycle. */
1193 static int add_internal_flows(struct ofproto_dpif *);
1195 static struct ofproto *
1198 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1199 return &ofproto->up;
1203 dealloc(struct ofproto *ofproto_)
1205 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1210 close_dpif_backer(struct dpif_backer *backer)
1212 struct shash_node *node;
1214 ovs_assert(backer->refcount > 0);
1216 if (--backer->refcount) {
1220 drop_key_clear(backer);
1221 hmap_destroy(&backer->drop_keys);
1223 simap_destroy(&backer->tnl_backers);
1224 hmap_destroy(&backer->odp_to_ofport_map);
1225 node = shash_find(&all_dpif_backers, backer->type);
1227 shash_delete(&all_dpif_backers, node);
1228 dpif_close(backer->dpif);
1233 /* Datapath port slated for removal from datapath. */
1234 struct odp_garbage {
1235 struct list list_node;
1240 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1242 struct dpif_backer *backer;
1243 struct dpif_port_dump port_dump;
1244 struct dpif_port port;
1245 struct shash_node *node;
1246 struct list garbage_list;
1247 struct odp_garbage *garbage, *next;
1253 backer = shash_find_data(&all_dpif_backers, type);
1260 backer_name = xasprintf("ovs-%s", type);
1262 /* Remove any existing datapaths, since we assume we're the only
1263 * userspace controlling the datapath. */
1265 dp_enumerate_names(type, &names);
1266 SSET_FOR_EACH(name, &names) {
1267 struct dpif *old_dpif;
1269 /* Don't remove our backer if it exists. */
1270 if (!strcmp(name, backer_name)) {
1274 if (dpif_open(name, type, &old_dpif)) {
1275 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1277 dpif_delete(old_dpif);
1278 dpif_close(old_dpif);
1281 sset_destroy(&names);
1283 backer = xmalloc(sizeof *backer);
1285 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1288 VLOG_ERR("failed to open datapath of type %s: %s", type,
1294 backer->type = xstrdup(type);
1295 backer->refcount = 1;
1296 hmap_init(&backer->odp_to_ofport_map);
1297 hmap_init(&backer->drop_keys);
1298 timer_set_duration(&backer->next_expiration, 1000);
1299 backer->need_revalidate = 0;
1300 simap_init(&backer->tnl_backers);
1301 tag_set_init(&backer->revalidate_set);
1304 dpif_flow_flush(backer->dpif);
1306 /* Loop through the ports already on the datapath and remove any
1307 * that we don't need anymore. */
1308 list_init(&garbage_list);
1309 dpif_port_dump_start(&port_dump, backer->dpif);
1310 while (dpif_port_dump_next(&port_dump, &port)) {
1311 node = shash_find(&init_ofp_ports, port.name);
1312 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1313 garbage = xmalloc(sizeof *garbage);
1314 garbage->odp_port = port.port_no;
1315 list_push_front(&garbage_list, &garbage->list_node);
1318 dpif_port_dump_done(&port_dump);
1320 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1321 dpif_port_del(backer->dpif, garbage->odp_port);
1322 list_remove(&garbage->list_node);
1326 shash_add(&all_dpif_backers, type, backer);
1328 error = dpif_recv_set(backer->dpif, true);
1330 VLOG_ERR("failed to listen on datapath of type %s: %s",
1331 type, strerror(error));
1332 close_dpif_backer(backer);
1340 construct(struct ofproto *ofproto_)
1342 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1343 struct shash_node *node, *next;
1348 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1353 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1354 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1356 ofproto->n_matches = 0;
1358 ofproto->netflow = NULL;
1359 ofproto->sflow = NULL;
1360 ofproto->ipfix = NULL;
1361 ofproto->stp = NULL;
1362 hmap_init(&ofproto->bundles);
1363 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1364 for (i = 0; i < MAX_MIRRORS; i++) {
1365 ofproto->mirrors[i] = NULL;
1367 ofproto->has_bonded_bundles = false;
1369 hmap_init(&ofproto->facets);
1370 hmap_init(&ofproto->subfacets);
1371 ofproto->governor = NULL;
1372 ofproto->consistency_rl = LLONG_MIN;
1374 for (i = 0; i < N_TABLES; i++) {
1375 struct table_dpif *table = &ofproto->tables[i];
1377 table->catchall_table = NULL;
1378 table->other_table = NULL;
1379 table->basis = random_uint32();
1382 list_init(&ofproto->completions);
1384 ofproto_dpif_unixctl_init();
1386 ofproto->has_mirrors = false;
1387 ofproto->has_bundle_action = false;
1389 hmap_init(&ofproto->vlandev_map);
1390 hmap_init(&ofproto->realdev_vid_map);
1392 sset_init(&ofproto->ports);
1393 sset_init(&ofproto->ghost_ports);
1394 sset_init(&ofproto->port_poll_set);
1395 ofproto->port_poll_errno = 0;
1397 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1398 struct iface_hint *iface_hint = node->data;
1400 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1401 /* Check if the datapath already has this port. */
1402 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1403 sset_add(&ofproto->ports, node->name);
1406 free(iface_hint->br_name);
1407 free(iface_hint->br_type);
1409 shash_delete(&init_ofp_ports, node);
1413 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1414 hash_string(ofproto->up.name, 0));
1415 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1417 ofproto_init_tables(ofproto_, N_TABLES);
1418 error = add_internal_flows(ofproto);
1419 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1422 ofproto->n_missed = 0;
1424 ofproto->max_n_subfacet = 0;
1425 ofproto->created = time_msec();
1426 ofproto->last_minute = ofproto->created;
1427 memset(&ofproto->hourly, 0, sizeof ofproto->hourly);
1428 memset(&ofproto->daily, 0, sizeof ofproto->daily);
1429 ofproto->subfacet_add_count = 0;
1430 ofproto->subfacet_del_count = 0;
1431 ofproto->total_subfacet_add_count = 0;
1432 ofproto->total_subfacet_del_count = 0;
1433 ofproto->total_subfacet_life_span = 0;
1434 ofproto->total_subfacet_count = 0;
1435 ofproto->n_update_stats = 0;
1441 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1442 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1444 struct ofputil_flow_mod fm;
1447 match_init_catchall(&fm.match);
1449 match_set_reg(&fm.match, 0, id);
1450 fm.new_cookie = htonll(0);
1451 fm.cookie = htonll(0);
1452 fm.cookie_mask = htonll(0);
1453 fm.table_id = TBL_INTERNAL;
1454 fm.command = OFPFC_ADD;
1455 fm.idle_timeout = 0;
1456 fm.hard_timeout = 0;
1460 fm.ofpacts = ofpacts->data;
1461 fm.ofpacts_len = ofpacts->size;
1463 error = ofproto_flow_mod(&ofproto->up, &fm);
1465 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1466 id, ofperr_to_string(error));
1470 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1471 ovs_assert(*rulep != NULL);
1477 add_internal_flows(struct ofproto_dpif *ofproto)
1479 struct ofpact_controller *controller;
1480 uint64_t ofpacts_stub[128 / 8];
1481 struct ofpbuf ofpacts;
1485 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1488 controller = ofpact_put_CONTROLLER(&ofpacts);
1489 controller->max_len = UINT16_MAX;
1490 controller->controller_id = 0;
1491 controller->reason = OFPR_NO_MATCH;
1492 ofpact_pad(&ofpacts);
1494 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1499 ofpbuf_clear(&ofpacts);
1500 error = add_internal_flow(ofproto, id++, &ofpacts,
1501 &ofproto->no_packet_in_rule);
1506 complete_operations(struct ofproto_dpif *ofproto)
1508 struct dpif_completion *c, *next;
1510 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1511 ofoperation_complete(c->op, 0);
1512 list_remove(&c->list_node);
1518 destruct(struct ofproto *ofproto_)
1520 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1521 struct rule_dpif *rule, *next_rule;
1522 struct oftable *table;
1525 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1526 complete_operations(ofproto);
1528 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1529 struct cls_cursor cursor;
1531 cls_cursor_init(&cursor, &table->cls, NULL);
1532 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1533 ofproto_rule_destroy(&rule->up);
1537 for (i = 0; i < MAX_MIRRORS; i++) {
1538 mirror_destroy(ofproto->mirrors[i]);
1541 netflow_destroy(ofproto->netflow);
1542 dpif_sflow_destroy(ofproto->sflow);
1543 hmap_destroy(&ofproto->bundles);
1544 mac_learning_destroy(ofproto->ml);
1546 hmap_destroy(&ofproto->facets);
1547 hmap_destroy(&ofproto->subfacets);
1548 governor_destroy(ofproto->governor);
1550 hmap_destroy(&ofproto->vlandev_map);
1551 hmap_destroy(&ofproto->realdev_vid_map);
1553 sset_destroy(&ofproto->ports);
1554 sset_destroy(&ofproto->ghost_ports);
1555 sset_destroy(&ofproto->port_poll_set);
1557 close_dpif_backer(ofproto->backer);
1561 run_fast(struct ofproto *ofproto_)
1563 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1564 struct ofport_dpif *ofport;
1566 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1567 port_run_fast(ofport);
1574 run(struct ofproto *ofproto_)
1576 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1577 struct ofport_dpif *ofport;
1578 struct ofbundle *bundle;
1582 complete_operations(ofproto);
1585 error = run_fast(ofproto_);
1590 if (ofproto->netflow) {
1591 if (netflow_run(ofproto->netflow)) {
1592 send_netflow_active_timeouts(ofproto);
1595 if (ofproto->sflow) {
1596 dpif_sflow_run(ofproto->sflow);
1599 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1602 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1607 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1609 /* Check the consistency of a random facet, to aid debugging. */
1610 if (time_msec() >= ofproto->consistency_rl
1611 && !hmap_is_empty(&ofproto->facets)
1612 && !ofproto->backer->need_revalidate) {
1613 struct facet *facet;
1615 ofproto->consistency_rl = time_msec() + 250;
1617 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1618 struct facet, hmap_node);
1619 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1621 if (!facet_check_consistency(facet)) {
1622 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1627 if (ofproto->governor) {
1630 governor_run(ofproto->governor);
1632 /* If the governor has shrunk to its minimum size and the number of
1633 * subfacets has dwindled, then drop the governor entirely.
1635 * For hysteresis, the number of subfacets to drop the governor is
1636 * smaller than the number needed to trigger its creation. */
1637 n_subfacets = hmap_count(&ofproto->subfacets);
1638 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1639 && governor_is_idle(ofproto->governor)) {
1640 governor_destroy(ofproto->governor);
1641 ofproto->governor = NULL;
1649 wait(struct ofproto *ofproto_)
1651 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1652 struct ofport_dpif *ofport;
1653 struct ofbundle *bundle;
1655 if (!clogged && !list_is_empty(&ofproto->completions)) {
1656 poll_immediate_wake();
1659 dpif_wait(ofproto->backer->dpif);
1660 dpif_recv_wait(ofproto->backer->dpif);
1661 if (ofproto->sflow) {
1662 dpif_sflow_wait(ofproto->sflow);
1664 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1665 poll_immediate_wake();
1667 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1670 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1671 bundle_wait(bundle);
1673 if (ofproto->netflow) {
1674 netflow_wait(ofproto->netflow);
1676 mac_learning_wait(ofproto->ml);
1678 if (ofproto->backer->need_revalidate) {
1679 /* Shouldn't happen, but if it does just go around again. */
1680 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1681 poll_immediate_wake();
1683 if (ofproto->governor) {
1684 governor_wait(ofproto->governor);
1689 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1691 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1693 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1694 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1698 flush(struct ofproto *ofproto_)
1700 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1701 struct subfacet *subfacet, *next_subfacet;
1702 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1706 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1707 &ofproto->subfacets) {
1708 if (subfacet->path != SF_NOT_INSTALLED) {
1709 batch[n_batch++] = subfacet;
1710 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1711 subfacet_destroy_batch(ofproto, batch, n_batch);
1715 subfacet_destroy(subfacet);
1720 subfacet_destroy_batch(ofproto, batch, n_batch);
1725 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1726 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1728 *arp_match_ip = true;
1729 *actions = (OFPUTIL_A_OUTPUT |
1730 OFPUTIL_A_SET_VLAN_VID |
1731 OFPUTIL_A_SET_VLAN_PCP |
1732 OFPUTIL_A_STRIP_VLAN |
1733 OFPUTIL_A_SET_DL_SRC |
1734 OFPUTIL_A_SET_DL_DST |
1735 OFPUTIL_A_SET_NW_SRC |
1736 OFPUTIL_A_SET_NW_DST |
1737 OFPUTIL_A_SET_NW_TOS |
1738 OFPUTIL_A_SET_TP_SRC |
1739 OFPUTIL_A_SET_TP_DST |
1744 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1746 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1747 struct dpif_dp_stats s;
1749 strcpy(ots->name, "classifier");
1751 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1753 ots->lookup_count = htonll(s.n_hit + s.n_missed);
1754 ots->matched_count = htonll(s.n_hit + ofproto->n_matches);
1757 static struct ofport *
1760 struct ofport_dpif *port = xmalloc(sizeof *port);
1765 port_dealloc(struct ofport *port_)
1767 struct ofport_dpif *port = ofport_dpif_cast(port_);
1772 port_construct(struct ofport *port_)
1774 struct ofport_dpif *port = ofport_dpif_cast(port_);
1775 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1776 const struct netdev *netdev = port->up.netdev;
1777 struct dpif_port dpif_port;
1780 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1781 port->bundle = NULL;
1784 port->tag = tag_create_random();
1785 port->may_enable = true;
1786 port->stp_port = NULL;
1787 port->stp_state = STP_DISABLED;
1788 port->tnl_port = NULL;
1789 hmap_init(&port->priorities);
1790 port->realdev_ofp_port = 0;
1791 port->vlandev_vid = 0;
1792 port->carrier_seq = netdev_get_carrier_resets(netdev);
1794 if (netdev_vport_is_patch(netdev)) {
1795 /* By bailing out here, we don't submit the port to the sFlow module
1796 * to be considered for counter polling export. This is correct
1797 * because the patch port represents an interface that sFlow considers
1798 * to be "internal" to the switch as a whole, and therefore not an
1799 * candidate for counter polling. */
1800 port->odp_port = OVSP_NONE;
1804 error = dpif_port_query_by_name(ofproto->backer->dpif,
1805 netdev_vport_get_dpif_port(netdev),
1811 port->odp_port = dpif_port.port_no;
1813 if (netdev_get_tunnel_config(netdev)) {
1814 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1816 /* Sanity-check that a mapping doesn't already exist. This
1817 * shouldn't happen for non-tunnel ports. */
1818 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1819 VLOG_ERR("port %s already has an OpenFlow port number",
1821 dpif_port_destroy(&dpif_port);
1825 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1826 hash_int(port->odp_port, 0));
1828 dpif_port_destroy(&dpif_port);
1830 if (ofproto->sflow) {
1831 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1838 port_destruct(struct ofport *port_)
1840 struct ofport_dpif *port = ofport_dpif_cast(port_);
1841 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1842 const char *dp_port_name = netdev_vport_get_dpif_port(port->up.netdev);
1843 const char *devname = netdev_get_name(port->up.netdev);
1845 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1846 /* The underlying device is still there, so delete it. This
1847 * happens when the ofproto is being destroyed, since the caller
1848 * assumes that removal of attached ports will happen as part of
1850 if (!port->tnl_port) {
1851 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1853 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1856 if (port->odp_port != OVSP_NONE && !port->tnl_port) {
1857 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1860 tnl_port_del(port->tnl_port);
1861 sset_find_and_delete(&ofproto->ports, devname);
1862 sset_find_and_delete(&ofproto->ghost_ports, devname);
1863 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1864 bundle_remove(port_);
1865 set_cfm(port_, NULL);
1866 if (ofproto->sflow) {
1867 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1870 ofport_clear_priorities(port);
1871 hmap_destroy(&port->priorities);
1875 port_modified(struct ofport *port_)
1877 struct ofport_dpif *port = ofport_dpif_cast(port_);
1879 if (port->bundle && port->bundle->bond) {
1880 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1885 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1887 struct ofport_dpif *port = ofport_dpif_cast(port_);
1888 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1889 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1891 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1892 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1893 OFPUTIL_PC_NO_PACKET_IN)) {
1894 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1896 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1897 bundle_update(port->bundle);
1903 set_sflow(struct ofproto *ofproto_,
1904 const struct ofproto_sflow_options *sflow_options)
1906 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1907 struct dpif_sflow *ds = ofproto->sflow;
1909 if (sflow_options) {
1911 struct ofport_dpif *ofport;
1913 ds = ofproto->sflow = dpif_sflow_create();
1914 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1915 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1917 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1919 dpif_sflow_set_options(ds, sflow_options);
1922 dpif_sflow_destroy(ds);
1923 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1924 ofproto->sflow = NULL;
1932 struct ofproto *ofproto_,
1933 const struct ofproto_ipfix_bridge_exporter_options *bridge_exporter_options,
1934 const struct ofproto_ipfix_flow_exporter_options *flow_exporters_options,
1935 size_t n_flow_exporters_options)
1937 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1938 struct dpif_ipfix *di = ofproto->ipfix;
1940 if (bridge_exporter_options || flow_exporters_options) {
1942 di = ofproto->ipfix = dpif_ipfix_create();
1944 dpif_ipfix_set_options(
1945 di, bridge_exporter_options, flow_exporters_options,
1946 n_flow_exporters_options);
1949 dpif_ipfix_destroy(di);
1950 ofproto->ipfix = NULL;
1957 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1959 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1966 struct ofproto_dpif *ofproto;
1968 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1969 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1970 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1973 if (cfm_configure(ofport->cfm, s)) {
1979 cfm_destroy(ofport->cfm);
1985 get_cfm_status(const struct ofport *ofport_,
1986 struct ofproto_cfm_status *status)
1988 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1991 status->faults = cfm_get_fault(ofport->cfm);
1992 status->remote_opstate = cfm_get_opup(ofport->cfm);
1993 status->health = cfm_get_health(ofport->cfm);
1994 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
2002 set_bfd(struct ofport *ofport_, const struct smap *cfg)
2004 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
2005 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2009 ofport->bfd = bfd_configure(old, netdev_get_name(ofport->up.netdev), cfg);
2010 if (ofport->bfd != old) {
2011 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2018 get_bfd_status(struct ofport *ofport_, struct smap *smap)
2020 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2023 bfd_get_status(ofport->bfd, smap);
2030 /* Spanning Tree. */
2033 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
2035 struct ofproto_dpif *ofproto = ofproto_;
2036 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
2037 struct ofport_dpif *ofport;
2039 ofport = stp_port_get_aux(sp);
2041 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
2042 ofproto->up.name, port_num);
2044 struct eth_header *eth = pkt->l2;
2046 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
2047 if (eth_addr_is_zero(eth->eth_src)) {
2048 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
2049 "with unknown MAC", ofproto->up.name, port_num);
2051 send_packet(ofport, pkt);
2057 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
2059 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
2061 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2063 /* Only revalidate flows if the configuration changed. */
2064 if (!s != !ofproto->stp) {
2065 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2069 if (!ofproto->stp) {
2070 ofproto->stp = stp_create(ofproto_->name, s->system_id,
2071 send_bpdu_cb, ofproto);
2072 ofproto->stp_last_tick = time_msec();
2075 stp_set_bridge_id(ofproto->stp, s->system_id);
2076 stp_set_bridge_priority(ofproto->stp, s->priority);
2077 stp_set_hello_time(ofproto->stp, s->hello_time);
2078 stp_set_max_age(ofproto->stp, s->max_age);
2079 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
2081 struct ofport *ofport;
2083 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
2084 set_stp_port(ofport, NULL);
2087 stp_destroy(ofproto->stp);
2088 ofproto->stp = NULL;
2095 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
2097 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2101 s->bridge_id = stp_get_bridge_id(ofproto->stp);
2102 s->designated_root = stp_get_designated_root(ofproto->stp);
2103 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
2112 update_stp_port_state(struct ofport_dpif *ofport)
2114 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2115 enum stp_state state;
2117 /* Figure out new state. */
2118 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
2122 if (ofport->stp_state != state) {
2123 enum ofputil_port_state of_state;
2126 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
2127 netdev_get_name(ofport->up.netdev),
2128 stp_state_name(ofport->stp_state),
2129 stp_state_name(state));
2130 if (stp_learn_in_state(ofport->stp_state)
2131 != stp_learn_in_state(state)) {
2132 /* xxx Learning action flows should also be flushed. */
2133 mac_learning_flush(ofproto->ml,
2134 &ofproto->backer->revalidate_set);
2136 fwd_change = stp_forward_in_state(ofport->stp_state)
2137 != stp_forward_in_state(state);
2139 ofproto->backer->need_revalidate = REV_STP;
2140 ofport->stp_state = state;
2141 ofport->stp_state_entered = time_msec();
2143 if (fwd_change && ofport->bundle) {
2144 bundle_update(ofport->bundle);
2147 /* Update the STP state bits in the OpenFlow port description. */
2148 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
2149 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
2150 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
2151 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
2152 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
2154 ofproto_port_set_state(&ofport->up, of_state);
2158 /* Configures STP on 'ofport_' using the settings defined in 's'. The
2159 * caller is responsible for assigning STP port numbers and ensuring
2160 * there are no duplicates. */
2162 set_stp_port(struct ofport *ofport_,
2163 const struct ofproto_port_stp_settings *s)
2165 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2166 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2167 struct stp_port *sp = ofport->stp_port;
2169 if (!s || !s->enable) {
2171 ofport->stp_port = NULL;
2172 stp_port_disable(sp);
2173 update_stp_port_state(ofport);
2176 } else if (sp && stp_port_no(sp) != s->port_num
2177 && ofport == stp_port_get_aux(sp)) {
2178 /* The port-id changed, so disable the old one if it's not
2179 * already in use by another port. */
2180 stp_port_disable(sp);
2183 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
2184 stp_port_enable(sp);
2186 stp_port_set_aux(sp, ofport);
2187 stp_port_set_priority(sp, s->priority);
2188 stp_port_set_path_cost(sp, s->path_cost);
2190 update_stp_port_state(ofport);
2196 get_stp_port_status(struct ofport *ofport_,
2197 struct ofproto_port_stp_status *s)
2199 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2200 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2201 struct stp_port *sp = ofport->stp_port;
2203 if (!ofproto->stp || !sp) {
2209 s->port_id = stp_port_get_id(sp);
2210 s->state = stp_port_get_state(sp);
2211 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2212 s->role = stp_port_get_role(sp);
2213 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2219 stp_run(struct ofproto_dpif *ofproto)
2222 long long int now = time_msec();
2223 long long int elapsed = now - ofproto->stp_last_tick;
2224 struct stp_port *sp;
2227 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2228 ofproto->stp_last_tick = now;
2230 while (stp_get_changed_port(ofproto->stp, &sp)) {
2231 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2234 update_stp_port_state(ofport);
2238 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2239 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2245 stp_wait(struct ofproto_dpif *ofproto)
2248 poll_timer_wait(1000);
2252 /* Returns true if STP should process 'flow'. */
2254 stp_should_process_flow(const struct flow *flow)
2256 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2260 stp_process_packet(const struct ofport_dpif *ofport,
2261 const struct ofpbuf *packet)
2263 struct ofpbuf payload = *packet;
2264 struct eth_header *eth = payload.data;
2265 struct stp_port *sp = ofport->stp_port;
2267 /* Sink packets on ports that have STP disabled when the bridge has
2269 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2273 /* Trim off padding on payload. */
2274 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2275 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2278 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2279 stp_received_bpdu(sp, payload.data, payload.size);
2283 static struct priority_to_dscp *
2284 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2286 struct priority_to_dscp *pdscp;
2289 hash = hash_int(priority, 0);
2290 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2291 if (pdscp->priority == priority) {
2299 ofport_clear_priorities(struct ofport_dpif *ofport)
2301 struct priority_to_dscp *pdscp, *next;
2303 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2304 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2310 set_queues(struct ofport *ofport_,
2311 const struct ofproto_port_queue *qdscp_list,
2314 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2315 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2316 struct hmap new = HMAP_INITIALIZER(&new);
2319 for (i = 0; i < n_qdscp; i++) {
2320 struct priority_to_dscp *pdscp;
2324 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2325 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2330 pdscp = get_priority(ofport, priority);
2332 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2334 pdscp = xmalloc(sizeof *pdscp);
2335 pdscp->priority = priority;
2337 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2340 if (pdscp->dscp != dscp) {
2342 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2345 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2348 if (!hmap_is_empty(&ofport->priorities)) {
2349 ofport_clear_priorities(ofport);
2350 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2353 hmap_swap(&new, &ofport->priorities);
2361 /* Expires all MAC learning entries associated with 'bundle' and forces its
2362 * ofproto to revalidate every flow.
2364 * Normally MAC learning entries are removed only from the ofproto associated
2365 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2366 * are removed from every ofproto. When patch ports and SLB bonds are in use
2367 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2368 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2369 * with the host from which it migrated. */
2371 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2373 struct ofproto_dpif *ofproto = bundle->ofproto;
2374 struct mac_learning *ml = ofproto->ml;
2375 struct mac_entry *mac, *next_mac;
2377 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2378 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2379 if (mac->port.p == bundle) {
2381 struct ofproto_dpif *o;
2383 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2385 struct mac_entry *e;
2387 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2390 mac_learning_expire(o->ml, e);
2396 mac_learning_expire(ml, mac);
2401 static struct ofbundle *
2402 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2404 struct ofbundle *bundle;
2406 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2407 &ofproto->bundles) {
2408 if (bundle->aux == aux) {
2415 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2416 * ones that are found to 'bundles'. */
2418 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2419 void **auxes, size_t n_auxes,
2420 struct hmapx *bundles)
2424 hmapx_init(bundles);
2425 for (i = 0; i < n_auxes; i++) {
2426 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2428 hmapx_add(bundles, bundle);
2434 bundle_update(struct ofbundle *bundle)
2436 struct ofport_dpif *port;
2438 bundle->floodable = true;
2439 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2440 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2441 || !stp_forward_in_state(port->stp_state)) {
2442 bundle->floodable = false;
2449 bundle_del_port(struct ofport_dpif *port)
2451 struct ofbundle *bundle = port->bundle;
2453 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2455 list_remove(&port->bundle_node);
2456 port->bundle = NULL;
2459 lacp_slave_unregister(bundle->lacp, port);
2462 bond_slave_unregister(bundle->bond, port);
2465 bundle_update(bundle);
2469 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
2470 struct lacp_slave_settings *lacp)
2472 struct ofport_dpif *port;
2474 port = get_ofp_port(bundle->ofproto, ofp_port);
2479 if (port->bundle != bundle) {
2480 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2482 bundle_del_port(port);
2485 port->bundle = bundle;
2486 list_push_back(&bundle->ports, &port->bundle_node);
2487 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2488 || !stp_forward_in_state(port->stp_state)) {
2489 bundle->floodable = false;
2493 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2494 lacp_slave_register(bundle->lacp, port, lacp);
2501 bundle_destroy(struct ofbundle *bundle)
2503 struct ofproto_dpif *ofproto;
2504 struct ofport_dpif *port, *next_port;
2511 ofproto = bundle->ofproto;
2512 for (i = 0; i < MAX_MIRRORS; i++) {
2513 struct ofmirror *m = ofproto->mirrors[i];
2515 if (m->out == bundle) {
2517 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2518 || hmapx_find_and_delete(&m->dsts, bundle)) {
2519 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2524 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2525 bundle_del_port(port);
2528 bundle_flush_macs(bundle, true);
2529 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2531 free(bundle->trunks);
2532 lacp_destroy(bundle->lacp);
2533 bond_destroy(bundle->bond);
2538 bundle_set(struct ofproto *ofproto_, void *aux,
2539 const struct ofproto_bundle_settings *s)
2541 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2542 bool need_flush = false;
2543 struct ofport_dpif *port;
2544 struct ofbundle *bundle;
2545 unsigned long *trunks;
2551 bundle_destroy(bundle_lookup(ofproto, aux));
2555 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2556 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2558 bundle = bundle_lookup(ofproto, aux);
2560 bundle = xmalloc(sizeof *bundle);
2562 bundle->ofproto = ofproto;
2563 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2564 hash_pointer(aux, 0));
2566 bundle->name = NULL;
2568 list_init(&bundle->ports);
2569 bundle->vlan_mode = PORT_VLAN_TRUNK;
2571 bundle->trunks = NULL;
2572 bundle->use_priority_tags = s->use_priority_tags;
2573 bundle->lacp = NULL;
2574 bundle->bond = NULL;
2576 bundle->floodable = true;
2578 bundle->src_mirrors = 0;
2579 bundle->dst_mirrors = 0;
2580 bundle->mirror_out = 0;
2583 if (!bundle->name || strcmp(s->name, bundle->name)) {
2585 bundle->name = xstrdup(s->name);
2590 if (!bundle->lacp) {
2591 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2592 bundle->lacp = lacp_create();
2594 lacp_configure(bundle->lacp, s->lacp);
2596 lacp_destroy(bundle->lacp);
2597 bundle->lacp = NULL;
2600 /* Update set of ports. */
2602 for (i = 0; i < s->n_slaves; i++) {
2603 if (!bundle_add_port(bundle, s->slaves[i],
2604 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2608 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2609 struct ofport_dpif *next_port;
2611 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2612 for (i = 0; i < s->n_slaves; i++) {
2613 if (s->slaves[i] == port->up.ofp_port) {
2618 bundle_del_port(port);
2622 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2624 if (list_is_empty(&bundle->ports)) {
2625 bundle_destroy(bundle);
2629 /* Set VLAN tagging mode */
2630 if (s->vlan_mode != bundle->vlan_mode
2631 || s->use_priority_tags != bundle->use_priority_tags) {
2632 bundle->vlan_mode = s->vlan_mode;
2633 bundle->use_priority_tags = s->use_priority_tags;
2638 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2639 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2641 if (vlan != bundle->vlan) {
2642 bundle->vlan = vlan;
2646 /* Get trunked VLANs. */
2647 switch (s->vlan_mode) {
2648 case PORT_VLAN_ACCESS:
2652 case PORT_VLAN_TRUNK:
2653 trunks = CONST_CAST(unsigned long *, s->trunks);
2656 case PORT_VLAN_NATIVE_UNTAGGED:
2657 case PORT_VLAN_NATIVE_TAGGED:
2658 if (vlan != 0 && (!s->trunks
2659 || !bitmap_is_set(s->trunks, vlan)
2660 || bitmap_is_set(s->trunks, 0))) {
2661 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2663 trunks = bitmap_clone(s->trunks, 4096);
2665 trunks = bitmap_allocate1(4096);
2667 bitmap_set1(trunks, vlan);
2668 bitmap_set0(trunks, 0);
2670 trunks = CONST_CAST(unsigned long *, s->trunks);
2677 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2678 free(bundle->trunks);
2679 if (trunks == s->trunks) {
2680 bundle->trunks = vlan_bitmap_clone(trunks);
2682 bundle->trunks = trunks;
2687 if (trunks != s->trunks) {
2692 if (!list_is_short(&bundle->ports)) {
2693 bundle->ofproto->has_bonded_bundles = true;
2695 if (bond_reconfigure(bundle->bond, s->bond)) {
2696 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2699 bundle->bond = bond_create(s->bond);
2700 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2703 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2704 bond_slave_register(bundle->bond, port, port->up.netdev);
2707 bond_destroy(bundle->bond);
2708 bundle->bond = NULL;
2711 /* If we changed something that would affect MAC learning, un-learn
2712 * everything on this port and force flow revalidation. */
2714 bundle_flush_macs(bundle, false);
2721 bundle_remove(struct ofport *port_)
2723 struct ofport_dpif *port = ofport_dpif_cast(port_);
2724 struct ofbundle *bundle = port->bundle;
2727 bundle_del_port(port);
2728 if (list_is_empty(&bundle->ports)) {
2729 bundle_destroy(bundle);
2730 } else if (list_is_short(&bundle->ports)) {
2731 bond_destroy(bundle->bond);
2732 bundle->bond = NULL;
2738 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2740 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2741 struct ofport_dpif *port = port_;
2742 uint8_t ea[ETH_ADDR_LEN];
2745 error = netdev_get_etheraddr(port->up.netdev, ea);
2747 struct ofpbuf packet;
2750 ofpbuf_init(&packet, 0);
2751 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2753 memcpy(packet_pdu, pdu, pdu_size);
2755 send_packet(port, &packet);
2756 ofpbuf_uninit(&packet);
2758 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2759 "%s (%s)", port->bundle->name,
2760 netdev_get_name(port->up.netdev), strerror(error));
2765 bundle_send_learning_packets(struct ofbundle *bundle)
2767 struct ofproto_dpif *ofproto = bundle->ofproto;
2768 int error, n_packets, n_errors;
2769 struct mac_entry *e;
2771 error = n_packets = n_errors = 0;
2772 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2773 if (e->port.p != bundle) {
2774 struct ofpbuf *learning_packet;
2775 struct ofport_dpif *port;
2779 /* The assignment to "port" is unnecessary but makes "grep"ing for
2780 * struct ofport_dpif more effective. */
2781 learning_packet = bond_compose_learning_packet(bundle->bond,
2785 ret = send_packet(port, learning_packet);
2786 ofpbuf_delete(learning_packet);
2796 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2797 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2798 "packets, last error was: %s",
2799 bundle->name, n_errors, n_packets, strerror(error));
2801 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2802 bundle->name, n_packets);
2807 bundle_run(struct ofbundle *bundle)
2810 lacp_run(bundle->lacp, send_pdu_cb);
2813 struct ofport_dpif *port;
2815 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2816 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2819 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2820 lacp_status(bundle->lacp));
2821 if (bond_should_send_learning_packets(bundle->bond)) {
2822 bundle_send_learning_packets(bundle);
2828 bundle_wait(struct ofbundle *bundle)
2831 lacp_wait(bundle->lacp);
2834 bond_wait(bundle->bond);
2841 mirror_scan(struct ofproto_dpif *ofproto)
2845 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2846 if (!ofproto->mirrors[idx]) {
2853 static struct ofmirror *
2854 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2858 for (i = 0; i < MAX_MIRRORS; i++) {
2859 struct ofmirror *mirror = ofproto->mirrors[i];
2860 if (mirror && mirror->aux == aux) {
2868 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2870 mirror_update_dups(struct ofproto_dpif *ofproto)
2874 for (i = 0; i < MAX_MIRRORS; i++) {
2875 struct ofmirror *m = ofproto->mirrors[i];
2878 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2882 for (i = 0; i < MAX_MIRRORS; i++) {
2883 struct ofmirror *m1 = ofproto->mirrors[i];
2890 for (j = i + 1; j < MAX_MIRRORS; j++) {
2891 struct ofmirror *m2 = ofproto->mirrors[j];
2893 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2894 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2895 m2->dup_mirrors |= m1->dup_mirrors;
2902 mirror_set(struct ofproto *ofproto_, void *aux,
2903 const struct ofproto_mirror_settings *s)
2905 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2906 mirror_mask_t mirror_bit;
2907 struct ofbundle *bundle;
2908 struct ofmirror *mirror;
2909 struct ofbundle *out;
2910 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2911 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2914 mirror = mirror_lookup(ofproto, aux);
2916 mirror_destroy(mirror);
2922 idx = mirror_scan(ofproto);
2924 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2926 ofproto->up.name, MAX_MIRRORS, s->name);
2930 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2931 mirror->ofproto = ofproto;
2934 mirror->out_vlan = -1;
2935 mirror->name = NULL;
2938 if (!mirror->name || strcmp(s->name, mirror->name)) {
2940 mirror->name = xstrdup(s->name);
2943 /* Get the new configuration. */
2944 if (s->out_bundle) {
2945 out = bundle_lookup(ofproto, s->out_bundle);
2947 mirror_destroy(mirror);
2953 out_vlan = s->out_vlan;
2955 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2956 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2958 /* If the configuration has not changed, do nothing. */
2959 if (hmapx_equals(&srcs, &mirror->srcs)
2960 && hmapx_equals(&dsts, &mirror->dsts)
2961 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2962 && mirror->out == out
2963 && mirror->out_vlan == out_vlan)
2965 hmapx_destroy(&srcs);
2966 hmapx_destroy(&dsts);
2970 hmapx_swap(&srcs, &mirror->srcs);
2971 hmapx_destroy(&srcs);
2973 hmapx_swap(&dsts, &mirror->dsts);
2974 hmapx_destroy(&dsts);
2976 free(mirror->vlans);
2977 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2980 mirror->out_vlan = out_vlan;
2982 /* Update bundles. */
2983 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2984 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2985 if (hmapx_contains(&mirror->srcs, bundle)) {
2986 bundle->src_mirrors |= mirror_bit;
2988 bundle->src_mirrors &= ~mirror_bit;
2991 if (hmapx_contains(&mirror->dsts, bundle)) {
2992 bundle->dst_mirrors |= mirror_bit;
2994 bundle->dst_mirrors &= ~mirror_bit;
2997 if (mirror->out == bundle) {
2998 bundle->mirror_out |= mirror_bit;
3000 bundle->mirror_out &= ~mirror_bit;
3004 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3005 ofproto->has_mirrors = true;
3006 mac_learning_flush(ofproto->ml,
3007 &ofproto->backer->revalidate_set);
3008 mirror_update_dups(ofproto);
3014 mirror_destroy(struct ofmirror *mirror)
3016 struct ofproto_dpif *ofproto;
3017 mirror_mask_t mirror_bit;
3018 struct ofbundle *bundle;
3025 ofproto = mirror->ofproto;
3026 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3027 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3029 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
3030 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
3031 bundle->src_mirrors &= ~mirror_bit;
3032 bundle->dst_mirrors &= ~mirror_bit;
3033 bundle->mirror_out &= ~mirror_bit;
3036 hmapx_destroy(&mirror->srcs);
3037 hmapx_destroy(&mirror->dsts);
3038 free(mirror->vlans);
3040 ofproto->mirrors[mirror->idx] = NULL;
3044 mirror_update_dups(ofproto);
3046 ofproto->has_mirrors = false;
3047 for (i = 0; i < MAX_MIRRORS; i++) {
3048 if (ofproto->mirrors[i]) {
3049 ofproto->has_mirrors = true;
3056 mirror_get_stats(struct ofproto *ofproto_, void *aux,
3057 uint64_t *packets, uint64_t *bytes)
3059 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3060 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
3063 *packets = *bytes = UINT64_MAX;
3069 *packets = mirror->packet_count;
3070 *bytes = mirror->byte_count;
3076 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
3078 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3079 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
3080 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
3086 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
3088 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3089 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
3090 return bundle && bundle->mirror_out != 0;
3094 forward_bpdu_changed(struct ofproto *ofproto_)
3096 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3097 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3101 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
3104 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3105 mac_learning_set_idle_time(ofproto->ml, idle_time);
3106 mac_learning_set_max_entries(ofproto->ml, max_entries);
3111 static struct ofport_dpif *
3112 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
3114 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
3115 return ofport ? ofport_dpif_cast(ofport) : NULL;
3118 static struct ofport_dpif *
3119 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
3121 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
3122 return port && &ofproto->up == port->up.ofproto ? port : NULL;
3126 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
3127 struct ofproto_port *ofproto_port,
3128 struct dpif_port *dpif_port)
3130 ofproto_port->name = dpif_port->name;
3131 ofproto_port->type = dpif_port->type;
3132 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
3135 static struct ofport_dpif *
3136 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
3138 const struct ofproto_dpif *ofproto;
3141 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
3146 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
3147 struct ofport *ofport;
3149 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
3150 if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) {
3151 return ofport_dpif_cast(ofport);
3158 port_run_fast(struct ofport_dpif *ofport)
3160 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
3161 struct ofpbuf packet;
3163 ofpbuf_init(&packet, 0);
3164 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
3165 send_packet(ofport, &packet);
3166 ofpbuf_uninit(&packet);
3169 if (ofport->bfd && bfd_should_send_packet(ofport->bfd)) {
3170 struct ofpbuf packet;
3172 ofpbuf_init(&packet, 0);
3173 bfd_put_packet(ofport->bfd, &packet, ofport->up.pp.hw_addr);
3174 send_packet(ofport, &packet);
3175 ofpbuf_uninit(&packet);
3180 port_run(struct ofport_dpif *ofport)
3182 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
3183 bool carrier_changed = carrier_seq != ofport->carrier_seq;
3184 bool enable = netdev_get_carrier(ofport->up.netdev);
3186 ofport->carrier_seq = carrier_seq;
3188 port_run_fast(ofport);
3190 if (ofport->tnl_port
3191 && tnl_port_reconfigure(&ofport->up, ofport->odp_port,
3192 &ofport->tnl_port)) {
3193 ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true;
3197 int cfm_opup = cfm_get_opup(ofport->cfm);
3199 cfm_run(ofport->cfm);
3200 enable = enable && !cfm_get_fault(ofport->cfm);
3202 if (cfm_opup >= 0) {
3203 enable = enable && cfm_opup;
3208 bfd_run(ofport->bfd);
3209 enable = enable && bfd_forwarding(ofport->bfd);
3212 if (ofport->bundle) {
3213 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3214 if (carrier_changed) {
3215 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3219 if (ofport->may_enable != enable) {
3220 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3222 if (ofproto->has_bundle_action) {
3223 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3227 ofport->may_enable = enable;
3231 port_wait(struct ofport_dpif *ofport)
3234 cfm_wait(ofport->cfm);
3238 bfd_wait(ofport->bfd);
3243 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3244 struct ofproto_port *ofproto_port)
3246 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3247 struct dpif_port dpif_port;
3250 if (sset_contains(&ofproto->ghost_ports, devname)) {
3251 const char *type = netdev_get_type_from_name(devname);
3253 /* We may be called before ofproto->up.port_by_name is populated with
3254 * the appropriate ofport. For this reason, we must get the name and
3255 * type from the netdev layer directly. */
3257 const struct ofport *ofport;
3259 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3260 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3261 ofproto_port->name = xstrdup(devname);
3262 ofproto_port->type = xstrdup(type);
3268 if (!sset_contains(&ofproto->ports, devname)) {
3271 error = dpif_port_query_by_name(ofproto->backer->dpif,
3272 devname, &dpif_port);
3274 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3280 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3282 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3283 const char *dp_port_name = netdev_vport_get_dpif_port(netdev);
3284 const char *devname = netdev_get_name(netdev);
3286 if (netdev_vport_is_patch(netdev)) {
3287 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3291 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3292 uint32_t port_no = UINT32_MAX;
3295 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3299 if (netdev_get_tunnel_config(netdev)) {
3300 simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no);
3304 if (netdev_get_tunnel_config(netdev)) {
3305 sset_add(&ofproto->ghost_ports, devname);
3307 sset_add(&ofproto->ports, devname);
3313 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
3315 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3316 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3323 sset_find_and_delete(&ofproto->ghost_ports,
3324 netdev_get_name(ofport->up.netdev));
3325 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3326 if (!ofport->tnl_port) {
3327 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3329 /* The caller is going to close ofport->up.netdev. If this is a
3330 * bonded port, then the bond is using that netdev, so remove it
3331 * from the bond. The client will need to reconfigure everything
3332 * after deleting ports, so then the slave will get re-added. */
3333 bundle_remove(&ofport->up);
3340 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3342 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3347 error = netdev_get_stats(ofport->up.netdev, stats);
3349 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3350 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3352 /* ofproto->stats.tx_packets represents packets that we created
3353 * internally and sent to some port (e.g. packets sent with
3354 * send_packet()). Account for them as if they had come from
3355 * OFPP_LOCAL and got forwarded. */
3357 if (stats->rx_packets != UINT64_MAX) {
3358 stats->rx_packets += ofproto->stats.tx_packets;
3361 if (stats->rx_bytes != UINT64_MAX) {
3362 stats->rx_bytes += ofproto->stats.tx_bytes;
3365 /* ofproto->stats.rx_packets represents packets that were received on
3366 * some port and we processed internally and dropped (e.g. STP).
3367 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3369 if (stats->tx_packets != UINT64_MAX) {
3370 stats->tx_packets += ofproto->stats.rx_packets;
3373 if (stats->tx_bytes != UINT64_MAX) {
3374 stats->tx_bytes += ofproto->stats.rx_bytes;
3381 /* Account packets for LOCAL port. */
3383 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
3384 size_t tx_size, size_t rx_size)
3386 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3389 ofproto->stats.rx_packets++;
3390 ofproto->stats.rx_bytes += rx_size;
3393 ofproto->stats.tx_packets++;
3394 ofproto->stats.tx_bytes += tx_size;
3398 struct port_dump_state {
3403 struct ofproto_port port;
3408 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3410 *statep = xzalloc(sizeof(struct port_dump_state));
3415 port_dump_next(const struct ofproto *ofproto_, void *state_,
3416 struct ofproto_port *port)
3418 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3419 struct port_dump_state *state = state_;
3420 const struct sset *sset;
3421 struct sset_node *node;
3423 if (state->has_port) {
3424 ofproto_port_destroy(&state->port);
3425 state->has_port = false;
3427 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3428 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3431 error = port_query_by_name(ofproto_, node->name, &state->port);
3433 *port = state->port;
3434 state->has_port = true;
3436 } else if (error != ENODEV) {
3441 if (!state->ghost) {
3442 state->ghost = true;
3445 return port_dump_next(ofproto_, state_, port);
3452 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3454 struct port_dump_state *state = state_;
3456 if (state->has_port) {
3457 ofproto_port_destroy(&state->port);
3464 port_poll(const struct ofproto *ofproto_, char **devnamep)
3466 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3468 if (ofproto->port_poll_errno) {
3469 int error = ofproto->port_poll_errno;
3470 ofproto->port_poll_errno = 0;
3474 if (sset_is_empty(&ofproto->port_poll_set)) {
3478 *devnamep = sset_pop(&ofproto->port_poll_set);
3483 port_poll_wait(const struct ofproto *ofproto_)
3485 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3486 dpif_port_poll_wait(ofproto->backer->dpif);
3490 port_is_lacp_current(const struct ofport *ofport_)
3492 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3493 return (ofport->bundle && ofport->bundle->lacp
3494 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3498 /* Upcall handling. */
3500 /* Flow miss batching.
3502 * Some dpifs implement operations faster when you hand them off in a batch.
3503 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3504 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3505 * more packets, plus possibly installing the flow in the dpif.
3507 * So far we only batch the operations that affect flow setup time the most.
3508 * It's possible to batch more than that, but the benefit might be minimal. */
3510 struct hmap_node hmap_node;
3511 struct ofproto_dpif *ofproto;
3513 enum odp_key_fitness key_fitness;
3514 const struct nlattr *key;
3516 struct initial_vals initial_vals;
3517 struct list packets;
3518 enum dpif_upcall_type upcall_type;
3519 uint32_t odp_in_port;
3522 struct flow_miss_op {
3523 struct dpif_op dpif_op;
3524 void *garbage; /* Pointer to pass to free(), NULL if none. */
3525 uint64_t stub[1024 / 8]; /* Temporary buffer. */
3528 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3529 * OpenFlow controller as necessary according to their individual
3530 * configurations. */
3532 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3533 const struct flow *flow)
3535 struct ofputil_packet_in pin;
3537 pin.packet = packet->data;
3538 pin.packet_len = packet->size;
3539 pin.reason = OFPR_NO_MATCH;
3540 pin.controller_id = 0;
3545 pin.send_len = 0; /* not used for flow table misses */
3547 flow_get_metadata(flow, &pin.fmd);
3549 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3552 static enum slow_path_reason
3553 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3554 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
3558 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3560 cfm_process_heartbeat(ofport->cfm, packet);
3563 } else if (ofport->bfd && bfd_should_process_flow(flow)) {
3565 bfd_process_packet(ofport->bfd, flow, packet);
3568 } else if (ofport->bundle && ofport->bundle->lacp
3569 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3571 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3574 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3576 stp_process_packet(ofport, packet);
3584 static struct flow_miss *
3585 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3586 const struct flow *flow, uint32_t hash)
3588 struct flow_miss *miss;
3590 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3591 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3599 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3600 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3601 * 'miss' is associated with a subfacet the caller must also initialize the
3602 * returned op->subfacet, and if anything needs to be freed after processing
3603 * the op, the caller must initialize op->garbage also. */
3605 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3606 struct flow_miss_op *op)
3608 if (miss->flow.vlan_tci != miss->initial_vals.vlan_tci) {
3609 /* This packet was received on a VLAN splinter port. We
3610 * added a VLAN to the packet to make the packet resemble
3611 * the flow, but the actions were composed assuming that
3612 * the packet contained no VLAN. So, we must remove the
3613 * VLAN header from the packet before trying to execute the
3615 eth_pop_vlan(packet);
3619 op->dpif_op.type = DPIF_OP_EXECUTE;
3620 op->dpif_op.u.execute.key = miss->key;
3621 op->dpif_op.u.execute.key_len = miss->key_len;
3622 op->dpif_op.u.execute.packet = packet;
3625 /* Helper for handle_flow_miss_without_facet() and
3626 * handle_flow_miss_with_facet(). */
3628 handle_flow_miss_common(struct rule_dpif *rule,
3629 struct ofpbuf *packet, const struct flow *flow)
3631 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3633 ofproto->n_matches++;
3635 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3637 * Extra-special case for fail-open mode.
3639 * We are in fail-open mode and the packet matched the fail-open
3640 * rule, but we are connected to a controller too. We should send
3641 * the packet up to the controller in the hope that it will try to
3642 * set up a flow and thereby allow us to exit fail-open.
3644 * See the top-level comment in fail-open.c for more information.
3646 send_packet_in_miss(ofproto, packet, flow);
3650 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3651 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3652 * installing a datapath flow. The answer is usually "yes" (a return value of
3653 * true). However, for short flows the cost of bookkeeping is much higher than
3654 * the benefits, so when the datapath holds a large number of flows we impose
3655 * some heuristics to decide which flows are likely to be worth tracking. */
3657 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3658 struct flow_miss *miss, uint32_t hash)
3660 if (!ofproto->governor) {
3663 n_subfacets = hmap_count(&ofproto->subfacets);
3664 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3668 ofproto->governor = governor_create(ofproto->up.name);
3671 return governor_should_install_flow(ofproto->governor, hash,
3672 list_size(&miss->packets));
3675 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3676 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3677 * increment '*n_ops'. */
3679 handle_flow_miss_without_facet(struct flow_miss *miss,
3680 struct rule_dpif *rule,
3681 struct flow_miss_op *ops, size_t *n_ops)
3683 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3684 long long int now = time_msec();
3685 struct action_xlate_ctx ctx;
3686 struct ofpbuf *packet;
3688 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3689 struct flow_miss_op *op = &ops[*n_ops];
3690 struct dpif_flow_stats stats;
3691 struct ofpbuf odp_actions;
3693 COVERAGE_INC(facet_suppress);
3695 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3697 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3698 rule_credit_stats(rule, &stats);
3700 action_xlate_ctx_init(&ctx, ofproto, &miss->flow,
3701 &miss->initial_vals, rule, 0, packet);
3702 ctx.resubmit_stats = &stats;
3703 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
3706 if (odp_actions.size) {
3707 struct dpif_execute *execute = &op->dpif_op.u.execute;
3709 init_flow_miss_execute_op(miss, packet, op);
3710 execute->actions = odp_actions.data;
3711 execute->actions_len = odp_actions.size;
3712 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3716 ofpbuf_uninit(&odp_actions);
3721 /* Handles 'miss', which matches 'facet'. May add any required datapath
3722 * operations to 'ops', incrementing '*n_ops' for each new op.
3724 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3725 * This is really important only for new facets: if we just called time_msec()
3726 * here, then the new subfacet or its packets could look (occasionally) as
3727 * though it was used some time after the facet was used. That can make a
3728 * one-packet flow look like it has a nonzero duration, which looks odd in
3729 * e.g. NetFlow statistics. */
3731 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3733 struct flow_miss_op *ops, size_t *n_ops)
3735 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3736 enum subfacet_path want_path;
3737 struct subfacet *subfacet;
3738 struct ofpbuf *packet;
3740 subfacet = subfacet_create(facet, miss, now);
3742 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3743 struct flow_miss_op *op = &ops[*n_ops];
3744 struct dpif_flow_stats stats;
3745 struct ofpbuf odp_actions;
3747 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3749 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3750 if (!subfacet->actions || subfacet->slow) {
3751 subfacet_make_actions(subfacet, packet, &odp_actions);
3754 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3755 subfacet_update_stats(subfacet, &stats);
3757 if (subfacet->actions_len) {
3758 struct dpif_execute *execute = &op->dpif_op.u.execute;
3760 init_flow_miss_execute_op(miss, packet, op);
3761 if (!subfacet->slow) {
3762 execute->actions = subfacet->actions;
3763 execute->actions_len = subfacet->actions_len;
3764 ofpbuf_uninit(&odp_actions);
3766 execute->actions = odp_actions.data;
3767 execute->actions_len = odp_actions.size;
3768 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3773 ofpbuf_uninit(&odp_actions);
3777 want_path = subfacet_want_path(subfacet->slow);
3778 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3779 struct flow_miss_op *op = &ops[(*n_ops)++];
3780 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3782 subfacet->path = want_path;
3785 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3786 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3787 put->key = miss->key;
3788 put->key_len = miss->key_len;
3789 if (want_path == SF_FAST_PATH) {
3790 put->actions = subfacet->actions;
3791 put->actions_len = subfacet->actions_len;
3793 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
3794 op->stub, sizeof op->stub,
3795 &put->actions, &put->actions_len);
3801 /* Handles flow miss 'miss'. May add any required datapath operations
3802 * to 'ops', incrementing '*n_ops' for each new op. */
3804 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3807 struct ofproto_dpif *ofproto = miss->ofproto;
3808 struct facet *facet;
3812 /* The caller must ensure that miss->hmap_node.hash contains
3813 * flow_hash(miss->flow, 0). */
3814 hash = miss->hmap_node.hash;
3816 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3818 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
3820 if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
3821 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
3825 facet = facet_create(rule, &miss->flow, hash);
3830 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3833 static struct drop_key *
3834 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3837 struct drop_key *drop_key;
3839 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3840 &backer->drop_keys) {
3841 if (drop_key->key_len == key_len
3842 && !memcmp(drop_key->key, key, key_len)) {
3850 drop_key_clear(struct dpif_backer *backer)
3852 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3853 struct drop_key *drop_key, *next;
3855 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3858 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3860 if (error && !VLOG_DROP_WARN(&rl)) {
3861 struct ds ds = DS_EMPTY_INITIALIZER;
3862 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3863 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3868 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3869 free(drop_key->key);
3874 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3875 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3876 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3877 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3878 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3879 * 'packet' ingressed.
3881 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3882 * 'flow''s in_port to OFPP_NONE.
3884 * This function does post-processing on data returned from
3885 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3886 * of the upcall processing logic. In particular, if the extracted in_port is
3887 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3888 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3889 * a VLAN header onto 'packet' (if it is nonnull).
3891 * Optionally, if 'initial_vals' is nonnull, sets 'initial_vals->vlan_tci'
3892 * to the VLAN TCI with which the packet was really received, that is, the
3893 * actual VLAN TCI extracted by odp_flow_key_to_flow(). (This differs from
3894 * the value returned in flow->vlan_tci only for packets received on
3895 * VLAN splinters.) Also, if received on an IP tunnel, sets
3896 * 'initial_vals->tunnel_ip_tos' to the tunnel's IP TOS.
3898 * Similarly, this function also includes some logic to help with tunnels. It
3899 * may modify 'flow' as necessary to make the tunneling implementation
3900 * transparent to the upcall processing logic.
3902 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3903 * or some other positive errno if there are other problems. */
3905 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3906 const struct nlattr *key, size_t key_len,
3907 struct flow *flow, enum odp_key_fitness *fitnessp,
3908 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3909 struct initial_vals *initial_vals)
3911 const struct ofport_dpif *port;
3912 enum odp_key_fitness fitness;
3915 fitness = odp_flow_key_to_flow(key, key_len, flow);
3916 if (fitness == ODP_FIT_ERROR) {
3922 initial_vals->vlan_tci = flow->vlan_tci;
3923 initial_vals->tunnel_ip_tos = flow->tunnel.ip_tos;
3927 *odp_in_port = flow->in_port;
3930 if (tnl_port_should_receive(flow)) {
3931 const struct ofport *ofport = tnl_port_receive(flow);
3933 flow->in_port = OFPP_NONE;
3936 port = ofport_dpif_cast(ofport);
3938 /* We can't reproduce 'key' from 'flow'. */
3939 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3941 /* XXX: Since the tunnel module is not scoped per backer, it's
3942 * theoretically possible that we'll receive an ofport belonging to an
3943 * entirely different datapath. In practice, this can't happen because
3944 * no platforms has two separate datapaths which each support
3946 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3948 port = odp_port_to_ofport(backer, flow->in_port);
3950 flow->in_port = OFPP_NONE;
3954 flow->in_port = port->up.ofp_port;
3955 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3957 /* Make the packet resemble the flow, so that it gets sent to
3958 * an OpenFlow controller properly, so that it looks correct
3959 * for sFlow, and so that flow_extract() will get the correct
3960 * vlan_tci if it is called on 'packet'.
3962 * The allocated space inside 'packet' probably also contains
3963 * 'key', that is, both 'packet' and 'key' are probably part of
3964 * a struct dpif_upcall (see the large comment on that
3965 * structure definition), so pushing data on 'packet' is in
3966 * general not a good idea since it could overwrite 'key' or
3967 * free it as a side effect. However, it's OK in this special
3968 * case because we know that 'packet' is inside a Netlink
3969 * attribute: pushing 4 bytes will just overwrite the 4-byte
3970 * "struct nlattr", which is fine since we don't need that
3971 * header anymore. */
3972 eth_push_vlan(packet, flow->vlan_tci);
3974 /* We can't reproduce 'key' from 'flow'. */
3975 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3981 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3986 *fitnessp = fitness;
3992 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3995 struct dpif_upcall *upcall;
3996 struct flow_miss *miss;
3997 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3998 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3999 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
4009 /* Construct the to-do list.
4011 * This just amounts to extracting the flow from each packet and sticking
4012 * the packets that have the same flow in the same "flow_miss" structure so
4013 * that we can process them together. */
4016 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
4017 struct flow_miss *miss = &misses[n_misses];
4018 struct flow_miss *existing_miss;
4019 struct ofproto_dpif *ofproto;
4020 uint32_t odp_in_port;
4025 error = ofproto_receive(backer, upcall->packet, upcall->key,
4026 upcall->key_len, &flow, &miss->key_fitness,
4027 &ofproto, &odp_in_port, &miss->initial_vals);
4028 if (error == ENODEV) {
4029 struct drop_key *drop_key;
4031 /* Received packet on port for which we couldn't associate
4032 * an ofproto. This can happen if a port is removed while
4033 * traffic is being received. Print a rate-limited message
4034 * in case it happens frequently. Install a drop flow so
4035 * that future packets of the flow are inexpensively dropped
4037 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
4040 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
4042 drop_key = xmalloc(sizeof *drop_key);
4043 drop_key->key = xmemdup(upcall->key, upcall->key_len);
4044 drop_key->key_len = upcall->key_len;
4046 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
4047 hash_bytes(drop_key->key, drop_key->key_len, 0));
4048 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
4049 drop_key->key, drop_key->key_len, NULL, 0, NULL);
4057 ofproto->n_missed++;
4058 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
4059 &flow.tunnel, flow.in_port, &miss->flow);
4061 /* Add other packets to a to-do list. */
4062 hash = flow_hash(&miss->flow, 0);
4063 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
4064 if (!existing_miss) {
4065 hmap_insert(&todo, &miss->hmap_node, hash);
4066 miss->ofproto = ofproto;
4067 miss->key = upcall->key;
4068 miss->key_len = upcall->key_len;
4069 miss->upcall_type = upcall->type;
4070 miss->odp_in_port = odp_in_port;
4071 list_init(&miss->packets);
4075 miss = existing_miss;
4077 list_push_back(&miss->packets, &upcall->packet->list_node);
4080 /* Process each element in the to-do list, constructing the set of
4081 * operations to batch. */
4083 HMAP_FOR_EACH (miss, hmap_node, &todo) {
4084 handle_flow_miss(miss, flow_miss_ops, &n_ops);
4086 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
4088 /* Execute batch. */
4089 for (i = 0; i < n_ops; i++) {
4090 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
4092 dpif_operate(backer->dpif, dpif_ops, n_ops);
4095 for (i = 0; i < n_ops; i++) {
4096 free(flow_miss_ops[i].garbage);
4098 hmap_destroy(&todo);
4101 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL, FLOW_SAMPLE_UPCALL,
4103 classify_upcall(const struct dpif_upcall *upcall)
4105 size_t userdata_len;
4106 union user_action_cookie cookie;
4108 /* First look at the upcall type. */
4109 switch (upcall->type) {
4110 case DPIF_UC_ACTION:
4116 case DPIF_N_UC_TYPES:
4118 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
4122 /* "action" upcalls need a closer look. */
4123 if (!upcall->userdata) {
4124 VLOG_WARN_RL(&rl, "action upcall missing cookie");
4127 userdata_len = nl_attr_get_size(upcall->userdata);
4128 if (userdata_len < sizeof cookie.type
4129 || userdata_len > sizeof cookie) {
4130 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
4134 memset(&cookie, 0, sizeof cookie);
4135 memcpy(&cookie, nl_attr_get(upcall->userdata), userdata_len);
4136 if (userdata_len == sizeof cookie.sflow
4137 && cookie.type == USER_ACTION_COOKIE_SFLOW) {
4138 return SFLOW_UPCALL;
4139 } else if (userdata_len == sizeof cookie.slow_path
4140 && cookie.type == USER_ACTION_COOKIE_SLOW_PATH) {
4142 } else if (userdata_len == sizeof cookie.flow_sample
4143 && cookie.type == USER_ACTION_COOKIE_FLOW_SAMPLE) {
4144 return FLOW_SAMPLE_UPCALL;
4145 } else if (userdata_len == sizeof cookie.ipfix
4146 && cookie.type == USER_ACTION_COOKIE_IPFIX) {
4147 return IPFIX_UPCALL;
4149 VLOG_WARN_RL(&rl, "invalid user cookie of type %"PRIu16
4150 " and size %zu", cookie.type, userdata_len);
4156 handle_sflow_upcall(struct dpif_backer *backer,
4157 const struct dpif_upcall *upcall)
4159 struct ofproto_dpif *ofproto;
4160 union user_action_cookie cookie;
4162 uint32_t odp_in_port;
4164 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4165 &flow, NULL, &ofproto, &odp_in_port, NULL)
4166 || !ofproto->sflow) {
4170 memset(&cookie, 0, sizeof cookie);
4171 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.sflow);
4172 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
4173 odp_in_port, &cookie);
4177 handle_flow_sample_upcall(struct dpif_backer *backer,
4178 const struct dpif_upcall *upcall)
4180 struct ofproto_dpif *ofproto;
4181 union user_action_cookie cookie;
4184 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4185 &flow, NULL, &ofproto, NULL, NULL)
4186 || !ofproto->ipfix) {
4190 memset(&cookie, 0, sizeof cookie);
4191 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof cookie.flow_sample);
4193 /* The flow reflects exactly the contents of the packet. Sample
4194 * the packet using it. */
4195 dpif_ipfix_flow_sample(ofproto->ipfix, upcall->packet, &flow,
4196 cookie.flow_sample.collector_set_id,
4197 cookie.flow_sample.probability,
4198 cookie.flow_sample.obs_domain_id,
4199 cookie.flow_sample.obs_point_id);
4203 handle_ipfix_upcall(struct dpif_backer *backer,
4204 const struct dpif_upcall *upcall)
4206 struct ofproto_dpif *ofproto;
4209 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
4210 &flow, NULL, &ofproto, NULL, NULL)
4211 || !ofproto->ipfix) {
4215 /* The flow reflects exactly the contents of the packet. Sample
4216 * the packet using it. */
4217 dpif_ipfix_bridge_sample(ofproto->ipfix, upcall->packet, &flow);
4221 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
4223 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
4224 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
4225 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
4230 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
4233 for (n_processed = 0; n_processed < max_batch; n_processed++) {
4234 struct dpif_upcall *upcall = &misses[n_misses];
4235 struct ofpbuf *buf = &miss_bufs[n_misses];
4238 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
4239 sizeof miss_buf_stubs[n_misses]);
4240 error = dpif_recv(backer->dpif, upcall, buf);
4246 switch (classify_upcall(upcall)) {
4248 /* Handle it later. */
4253 handle_sflow_upcall(backer, upcall);
4257 case FLOW_SAMPLE_UPCALL:
4258 handle_flow_sample_upcall(backer, upcall);
4263 handle_ipfix_upcall(backer, upcall);
4273 /* Handle deferred MISS_UPCALL processing. */
4274 handle_miss_upcalls(backer, misses, n_misses);
4275 for (i = 0; i < n_misses; i++) {
4276 ofpbuf_uninit(&miss_bufs[i]);
4282 /* Flow expiration. */
4284 static int subfacet_max_idle(const struct ofproto_dpif *);
4285 static void update_stats(struct dpif_backer *);
4286 static void rule_expire(struct rule_dpif *);
4287 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
4289 /* This function is called periodically by run(). Its job is to collect
4290 * updates for the flows that have been installed into the datapath, most
4291 * importantly when they last were used, and then use that information to
4292 * expire flows that have not been used recently.
4294 * Returns the number of milliseconds after which it should be called again. */
4296 expire(struct dpif_backer *backer)
4298 struct ofproto_dpif *ofproto;
4299 int max_idle = INT32_MAX;
4301 /* Periodically clear out the drop keys in an effort to keep them
4302 * relatively few. */
4303 drop_key_clear(backer);
4305 /* Update stats for each flow in the backer. */
4306 update_stats(backer);
4308 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4309 struct rule *rule, *next_rule;
4312 if (ofproto->backer != backer) {
4316 /* Keep track of the max number of flows per ofproto_dpif. */
4317 update_max_subfacet_count(ofproto);
4319 /* Expire subfacets that have been idle too long. */
4320 dp_max_idle = subfacet_max_idle(ofproto);
4321 expire_subfacets(ofproto, dp_max_idle);
4323 max_idle = MIN(max_idle, dp_max_idle);
4325 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4327 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4328 &ofproto->up.expirable) {
4329 rule_expire(rule_dpif_cast(rule));
4332 /* All outstanding data in existing flows has been accounted, so it's a
4333 * good time to do bond rebalancing. */
4334 if (ofproto->has_bonded_bundles) {
4335 struct ofbundle *bundle;
4337 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4339 bond_rebalance(bundle->bond, &backer->revalidate_set);
4345 return MIN(max_idle, 1000);
4348 /* Updates flow table statistics given that the datapath just reported 'stats'
4349 * as 'subfacet''s statistics. */
4351 update_subfacet_stats(struct subfacet *subfacet,
4352 const struct dpif_flow_stats *stats)
4354 struct facet *facet = subfacet->facet;
4356 if (stats->n_packets >= subfacet->dp_packet_count) {
4357 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
4358 facet->packet_count += extra;
4360 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4363 if (stats->n_bytes >= subfacet->dp_byte_count) {
4364 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
4366 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4369 subfacet->dp_packet_count = stats->n_packets;
4370 subfacet->dp_byte_count = stats->n_bytes;
4372 facet->tcp_flags |= stats->tcp_flags;
4374 subfacet_update_time(subfacet, stats->used);
4375 if (facet->accounted_bytes < facet->byte_count) {
4377 facet_account(facet);
4378 facet->accounted_bytes = facet->byte_count;
4382 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4383 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4385 delete_unexpected_flow(struct ofproto_dpif *ofproto,
4386 const struct nlattr *key, size_t key_len)
4388 if (!VLOG_DROP_WARN(&rl)) {
4392 odp_flow_key_format(key, key_len, &s);
4393 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
4397 COVERAGE_INC(facet_unexpected);
4398 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
4401 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4403 * This function also pushes statistics updates to rules which each facet
4404 * resubmits into. Generally these statistics will be accurate. However, if a
4405 * facet changes the rule it resubmits into at some time in between
4406 * update_stats() runs, it is possible that statistics accrued to the
4407 * old rule will be incorrectly attributed to the new rule. This could be
4408 * avoided by calling update_stats() whenever rules are created or
4409 * deleted. However, the performance impact of making so many calls to the
4410 * datapath do not justify the benefit of having perfectly accurate statistics.
4412 * In addition, this function maintains per ofproto flow hit counts. The patch
4413 * port is not treated specially. e.g. A packet ingress from br0 patched into
4414 * br1 will increase the hit count of br0 by 1, however, does not affect
4415 * the hit or miss counts of br1.
4418 update_stats(struct dpif_backer *backer)
4420 const struct dpif_flow_stats *stats;
4421 struct dpif_flow_dump dump;
4422 const struct nlattr *key;
4423 struct ofproto_dpif *ofproto;
4426 dpif_flow_dump_start(&dump, backer->dpif);
4427 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4429 struct subfacet *subfacet;
4430 struct ofport_dpif *ofport;
4433 if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto,
4438 ofproto->total_subfacet_count += hmap_count(&ofproto->subfacets);
4439 ofproto->n_update_stats++;
4441 ofport = get_ofp_port(ofproto, flow.in_port);
4442 if (ofport && ofport->tnl_port) {
4443 netdev_vport_inc_rx(ofport->up.netdev, stats);
4446 key_hash = odp_flow_key_hash(key, key_len);
4447 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
4448 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4450 /* Update ofproto_dpif's hit count. */
4451 if (stats->n_packets > subfacet->dp_packet_count) {
4452 uint64_t delta = stats->n_packets - subfacet->dp_packet_count;
4453 dpif_stats_update_hit_count(ofproto, delta);
4456 update_subfacet_stats(subfacet, stats);
4460 /* Stats are updated per-packet. */
4463 case SF_NOT_INSTALLED:
4465 delete_unexpected_flow(ofproto, key, key_len);
4470 dpif_flow_dump_done(&dump);
4472 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4473 update_moving_averages(ofproto);
4478 /* Calculates and returns the number of milliseconds of idle time after which
4479 * subfacets should expire from the datapath. When a subfacet expires, we fold
4480 * its statistics into its facet, and when a facet's last subfacet expires, we
4481 * fold its statistic into its rule. */
4483 subfacet_max_idle(const struct ofproto_dpif *ofproto)
4486 * Idle time histogram.
4488 * Most of the time a switch has a relatively small number of subfacets.
4489 * When this is the case we might as well keep statistics for all of them
4490 * in userspace and to cache them in the kernel datapath for performance as
4493 * As the number of subfacets increases, the memory required to maintain
4494 * statistics about them in userspace and in the kernel becomes
4495 * significant. However, with a large number of subfacets it is likely
4496 * that only a few of them are "heavy hitters" that consume a large amount
4497 * of bandwidth. At this point, only heavy hitters are worth caching in
4498 * the kernel and maintaining in userspaces; other subfacets we can
4501 * The technique used to compute the idle time is to build a histogram with
4502 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4503 * that is installed in the kernel gets dropped in the appropriate bucket.
4504 * After the histogram has been built, we compute the cutoff so that only
4505 * the most-recently-used 1% of subfacets (but at least
4506 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
4507 * the most-recently-used bucket of subfacets is kept, so actually an
4508 * arbitrary number of subfacets can be kept in any given expiration run
4509 * (though the next run will delete most of those unless they receive
4512 * This requires a second pass through the subfacets, in addition to the
4513 * pass made by update_stats(), because the former function never looks at
4514 * uninstallable subfacets.
4516 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4517 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4518 int buckets[N_BUCKETS] = { 0 };
4519 int total, subtotal, bucket;
4520 struct subfacet *subfacet;
4524 total = hmap_count(&ofproto->subfacets);
4525 if (total <= ofproto->up.flow_eviction_threshold) {
4526 return N_BUCKETS * BUCKET_WIDTH;
4529 /* Build histogram. */
4531 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
4532 long long int idle = now - subfacet->used;
4533 int bucket = (idle <= 0 ? 0
4534 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4535 : (unsigned int) idle / BUCKET_WIDTH);
4539 /* Find the first bucket whose flows should be expired. */
4540 subtotal = bucket = 0;
4542 subtotal += buckets[bucket++];
4543 } while (bucket < N_BUCKETS &&
4544 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
4546 if (VLOG_IS_DBG_ENABLED()) {
4550 ds_put_cstr(&s, "keep");
4551 for (i = 0; i < N_BUCKETS; i++) {
4553 ds_put_cstr(&s, ", drop");
4556 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4559 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
4563 return bucket * BUCKET_WIDTH;
4567 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
4569 /* Cutoff time for most flows. */
4570 long long int normal_cutoff = time_msec() - dp_max_idle;
4572 /* We really want to keep flows for special protocols around, so use a more
4573 * conservative cutoff. */
4574 long long int special_cutoff = time_msec() - 10000;
4576 struct subfacet *subfacet, *next_subfacet;
4577 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4581 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4582 &ofproto->subfacets) {
4583 long long int cutoff;
4585 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)
4588 if (subfacet->used < cutoff) {
4589 if (subfacet->path != SF_NOT_INSTALLED) {
4590 batch[n_batch++] = subfacet;
4591 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4592 subfacet_destroy_batch(ofproto, batch, n_batch);
4596 subfacet_destroy(subfacet);
4602 subfacet_destroy_batch(ofproto, batch, n_batch);
4606 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4607 * then delete it entirely. */
4609 rule_expire(struct rule_dpif *rule)
4611 struct facet *facet, *next_facet;
4615 if (rule->up.pending) {
4616 /* We'll have to expire it later. */
4620 /* Has 'rule' expired? */
4622 if (rule->up.hard_timeout
4623 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4624 reason = OFPRR_HARD_TIMEOUT;
4625 } else if (rule->up.idle_timeout
4626 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4627 reason = OFPRR_IDLE_TIMEOUT;
4632 COVERAGE_INC(ofproto_dpif_expired);
4634 /* Update stats. (This is a no-op if the rule expired due to an idle
4635 * timeout, because that only happens when the rule has no facets left.) */
4636 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4637 facet_remove(facet);
4640 /* Get rid of the rule. */
4641 ofproto_rule_expire(&rule->up, reason);
4646 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
4648 * The caller must already have determined that no facet with an identical
4649 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
4650 * the ofproto's classifier table.
4652 * 'hash' must be the return value of flow_hash(flow, 0).
4654 * The facet will initially have no subfacets. The caller should create (at
4655 * least) one subfacet with subfacet_create(). */
4656 static struct facet *
4657 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
4659 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4660 struct facet *facet;
4662 facet = xzalloc(sizeof *facet);
4663 facet->used = time_msec();
4664 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4665 list_push_back(&rule->facets, &facet->list_node);
4667 facet->flow = *flow;
4668 list_init(&facet->subfacets);
4669 netflow_flow_init(&facet->nf_flow);
4670 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4672 facet->learn_rl = time_msec() + 500;
4678 facet_free(struct facet *facet)
4683 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4684 * 'packet', which arrived on 'in_port'. */
4686 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4687 const struct nlattr *odp_actions, size_t actions_len,
4688 struct ofpbuf *packet)
4690 struct odputil_keybuf keybuf;
4694 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4695 odp_flow_key_from_flow(&key, flow,
4696 ofp_port_to_odp_port(ofproto, flow->in_port));
4698 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4699 odp_actions, actions_len, packet);
4703 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4705 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4706 * rule's statistics, via subfacet_uninstall().
4708 * - Removes 'facet' from its rule and from ofproto->facets.
4711 facet_remove(struct facet *facet)
4713 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4714 struct subfacet *subfacet, *next_subfacet;
4716 ovs_assert(!list_is_empty(&facet->subfacets));
4718 /* First uninstall all of the subfacets to get final statistics. */
4719 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4720 subfacet_uninstall(subfacet);
4723 /* Flush the final stats to the rule.
4725 * This might require us to have at least one subfacet around so that we
4726 * can use its actions for accounting in facet_account(), which is why we
4727 * have uninstalled but not yet destroyed the subfacets. */
4728 facet_flush_stats(facet);
4730 /* Now we're really all done so destroy everything. */
4731 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4732 &facet->subfacets) {
4733 subfacet_destroy__(subfacet);
4735 hmap_remove(&ofproto->facets, &facet->hmap_node);
4736 list_remove(&facet->list_node);
4740 /* Feed information from 'facet' back into the learning table to keep it in
4741 * sync with what is actually flowing through the datapath. */
4743 facet_learn(struct facet *facet)
4745 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4746 struct subfacet *subfacet= CONTAINER_OF(list_front(&facet->subfacets),
4747 struct subfacet, list_node);
4748 long long int now = time_msec();
4749 struct action_xlate_ctx ctx;
4751 if (!facet->has_fin_timeout && now < facet->learn_rl) {
4755 facet->learn_rl = now + 500;
4757 if (!facet->has_learn
4758 && !facet->has_normal
4759 && (!facet->has_fin_timeout
4760 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4764 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4765 &subfacet->initial_vals,
4766 facet->rule, facet->tcp_flags, NULL);
4767 ctx.may_learn = true;
4768 xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
4769 facet->rule->up.ofpacts_len);
4773 facet_account(struct facet *facet)
4775 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4776 struct subfacet *subfacet = facet_get_subfacet(facet);
4777 const struct nlattr *a;
4782 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
4785 n_bytes = facet->byte_count - facet->accounted_bytes;
4787 /* This loop feeds byte counters to bond_account() for rebalancing to use
4788 * as a basis. We also need to track the actual VLAN on which the packet
4789 * is going to be sent to ensure that it matches the one passed to
4790 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4793 * We use the actions from an arbitrary subfacet because they should all
4794 * be equally valid for our purpose. */
4795 vlan_tci = facet->flow.vlan_tci;
4796 NL_ATTR_FOR_EACH_UNSAFE (a, left,
4797 subfacet->actions, subfacet->actions_len) {
4798 const struct ovs_action_push_vlan *vlan;
4799 struct ofport_dpif *port;
4801 switch (nl_attr_type(a)) {
4802 case OVS_ACTION_ATTR_OUTPUT:
4803 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4804 if (port && port->bundle && port->bundle->bond) {
4805 bond_account(port->bundle->bond, &facet->flow,
4806 vlan_tci_to_vid(vlan_tci), n_bytes);
4810 case OVS_ACTION_ATTR_POP_VLAN:
4811 vlan_tci = htons(0);
4814 case OVS_ACTION_ATTR_PUSH_VLAN:
4815 vlan = nl_attr_get(a);
4816 vlan_tci = vlan->vlan_tci;
4822 /* Returns true if the only action for 'facet' is to send to the controller.
4823 * (We don't report NetFlow expiration messages for such facets because they
4824 * are just part of the control logic for the network, not real traffic). */
4826 facet_is_controller_flow(struct facet *facet)
4829 const struct rule *rule = &facet->rule->up;
4830 const struct ofpact *ofpacts = rule->ofpacts;
4831 size_t ofpacts_len = rule->ofpacts_len;
4833 if (ofpacts_len > 0 &&
4834 ofpacts->type == OFPACT_CONTROLLER &&
4835 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4842 /* Folds all of 'facet''s statistics into its rule. Also updates the
4843 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4844 * 'facet''s statistics in the datapath should have been zeroed and folded into
4845 * its packet and byte counts before this function is called. */
4847 facet_flush_stats(struct facet *facet)
4849 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4850 struct subfacet *subfacet;
4852 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4853 ovs_assert(!subfacet->dp_byte_count);
4854 ovs_assert(!subfacet->dp_packet_count);
4857 facet_push_stats(facet);
4858 if (facet->accounted_bytes < facet->byte_count) {
4859 facet_account(facet);
4860 facet->accounted_bytes = facet->byte_count;
4863 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4864 struct ofexpired expired;
4865 expired.flow = facet->flow;
4866 expired.packet_count = facet->packet_count;
4867 expired.byte_count = facet->byte_count;
4868 expired.used = facet->used;
4869 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4872 facet->rule->packet_count += facet->packet_count;
4873 facet->rule->byte_count += facet->byte_count;
4875 /* Reset counters to prevent double counting if 'facet' ever gets
4877 facet_reset_counters(facet);
4879 netflow_flow_clear(&facet->nf_flow);
4880 facet->tcp_flags = 0;
4883 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4884 * Returns it if found, otherwise a null pointer.
4886 * 'hash' must be the return value of flow_hash(flow, 0).
4888 * The returned facet might need revalidation; use facet_lookup_valid()
4889 * instead if that is important. */
4890 static struct facet *
4891 facet_find(struct ofproto_dpif *ofproto,
4892 const struct flow *flow, uint32_t hash)
4894 struct facet *facet;
4896 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4897 if (flow_equal(flow, &facet->flow)) {
4905 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4906 * Returns it if found, otherwise a null pointer.
4908 * 'hash' must be the return value of flow_hash(flow, 0).
4910 * The returned facet is guaranteed to be valid. */
4911 static struct facet *
4912 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4915 struct facet *facet;
4917 facet = facet_find(ofproto, flow, hash);
4919 && (ofproto->backer->need_revalidate
4920 || tag_set_intersects(&ofproto->backer->revalidate_set,
4922 facet_revalidate(facet);
4924 /* facet_revalidate() may have destroyed 'facet'. */
4925 facet = facet_find(ofproto, flow, hash);
4931 /* Return a subfacet from 'facet'. A facet consists of one or more
4932 * subfacets, and this function returns one of them. */
4933 static struct subfacet *facet_get_subfacet(struct facet *facet)
4935 return CONTAINER_OF(list_front(&facet->subfacets), struct subfacet,
4940 subfacet_path_to_string(enum subfacet_path path)
4943 case SF_NOT_INSTALLED:
4944 return "not installed";
4946 return "in fast path";
4948 return "in slow path";
4954 /* Returns the path in which a subfacet should be installed if its 'slow'
4955 * member has the specified value. */
4956 static enum subfacet_path
4957 subfacet_want_path(enum slow_path_reason slow)
4959 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
4962 /* Returns true if 'subfacet' needs to have its datapath flow updated,
4963 * supposing that its actions have been recalculated as 'want_actions' and that
4964 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
4966 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
4967 const struct ofpbuf *want_actions)
4969 enum subfacet_path want_path = subfacet_want_path(slow);
4970 return (want_path != subfacet->path
4971 || (want_path == SF_FAST_PATH
4972 && (subfacet->actions_len != want_actions->size
4973 || memcmp(subfacet->actions, want_actions->data,
4974 subfacet->actions_len))));
4978 facet_check_consistency(struct facet *facet)
4980 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4982 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4984 uint64_t odp_actions_stub[1024 / 8];
4985 struct ofpbuf odp_actions;
4987 struct rule_dpif *rule;
4988 struct subfacet *subfacet;
4989 bool may_log = false;
4992 /* Check the rule for consistency. */
4993 rule = rule_dpif_lookup(ofproto, &facet->flow);
4994 ok = rule == facet->rule;
4996 may_log = !VLOG_DROP_WARN(&rl);
5001 flow_format(&s, &facet->flow);
5002 ds_put_format(&s, ": facet associated with wrong rule (was "
5003 "table=%"PRIu8",", facet->rule->up.table_id);
5004 cls_rule_format(&facet->rule->up.cr, &s);
5005 ds_put_format(&s, ") (should have been table=%"PRIu8",",
5007 cls_rule_format(&rule->up.cr, &s);
5008 ds_put_char(&s, ')');
5010 VLOG_WARN("%s", ds_cstr(&s));
5015 /* Check the datapath actions for consistency. */
5016 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5017 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5018 enum subfacet_path want_path;
5019 struct action_xlate_ctx ctx;
5022 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
5023 &subfacet->initial_vals, rule, 0, NULL);
5024 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
5027 if (subfacet->path == SF_NOT_INSTALLED) {
5028 /* This only happens if the datapath reported an error when we
5029 * tried to install the flow. Don't flag another error here. */
5033 want_path = subfacet_want_path(subfacet->slow);
5034 if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) {
5035 /* The actions for slow-path flows may legitimately vary from one
5036 * packet to the next. We're done. */
5040 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
5044 /* Inconsistency! */
5046 may_log = !VLOG_DROP_WARN(&rl);
5050 /* Rate-limited, skip reporting. */
5055 odp_flow_key_format(subfacet->key, subfacet->key_len, &s);
5057 ds_put_cstr(&s, ": inconsistency in subfacet");
5058 if (want_path != subfacet->path) {
5059 enum odp_key_fitness fitness = subfacet->key_fitness;
5061 ds_put_format(&s, " (%s, fitness=%s)",
5062 subfacet_path_to_string(subfacet->path),
5063 odp_key_fitness_to_string(fitness));
5064 ds_put_format(&s, " (should have been %s)",
5065 subfacet_path_to_string(want_path));
5066 } else if (want_path == SF_FAST_PATH) {
5067 ds_put_cstr(&s, " (actions were: ");
5068 format_odp_actions(&s, subfacet->actions,
5069 subfacet->actions_len);
5070 ds_put_cstr(&s, ") (correct actions: ");
5071 format_odp_actions(&s, odp_actions.data, odp_actions.size);
5072 ds_put_char(&s, ')');
5074 ds_put_cstr(&s, " (actions: ");
5075 format_odp_actions(&s, subfacet->actions,
5076 subfacet->actions_len);
5077 ds_put_char(&s, ')');
5079 VLOG_WARN("%s", ds_cstr(&s));
5082 ofpbuf_uninit(&odp_actions);
5087 /* Re-searches the classifier for 'facet':
5089 * - If the rule found is different from 'facet''s current rule, moves
5090 * 'facet' to the new rule and recompiles its actions.
5092 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
5093 * where it is and recompiles its actions anyway.
5095 * - If any of 'facet''s subfacets correspond to a new flow according to
5096 * ofproto_receive(), 'facet' is removed. */
5098 facet_revalidate(struct facet *facet)
5100 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5102 struct nlattr *odp_actions;
5105 struct actions *new_actions;
5107 struct action_xlate_ctx ctx;
5108 uint64_t odp_actions_stub[1024 / 8];
5109 struct ofpbuf odp_actions;
5111 struct rule_dpif *new_rule;
5112 struct subfacet *subfacet;
5115 COVERAGE_INC(facet_revalidate);
5117 /* Check that child subfacets still correspond to this facet. Tunnel
5118 * configuration changes could cause a subfacet's OpenFlow in_port to
5120 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5121 struct ofproto_dpif *recv_ofproto;
5122 struct flow recv_flow;
5125 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
5126 subfacet->key_len, &recv_flow, NULL,
5127 &recv_ofproto, NULL, NULL);
5129 || recv_ofproto != ofproto
5130 || memcmp(&recv_flow, &facet->flow, sizeof recv_flow)) {
5131 facet_remove(facet);
5136 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
5138 /* Calculate new datapath actions.
5140 * We do not modify any 'facet' state yet, because we might need to, e.g.,
5141 * emit a NetFlow expiration and, if so, we need to have the old state
5142 * around to properly compose it. */
5144 /* If the datapath actions changed or the installability changed,
5145 * then we need to talk to the datapath. */
5148 memset(&ctx, 0, sizeof ctx);
5149 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5150 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5151 enum slow_path_reason slow;
5153 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
5154 &subfacet->initial_vals, new_rule, 0, NULL);
5155 xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len,
5158 slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
5159 if (subfacet_should_install(subfacet, slow, &odp_actions)) {
5160 struct dpif_flow_stats stats;
5162 subfacet_install(subfacet,
5163 odp_actions.data, odp_actions.size, &stats, slow);
5164 subfacet_update_stats(subfacet, &stats);
5167 new_actions = xcalloc(list_size(&facet->subfacets),
5168 sizeof *new_actions);
5170 new_actions[i].odp_actions = xmemdup(odp_actions.data,
5172 new_actions[i].actions_len = odp_actions.size;
5177 ofpbuf_uninit(&odp_actions);
5180 facet_flush_stats(facet);
5183 /* Update 'facet' now that we've taken care of all the old state. */
5184 facet->tags = ctx.tags;
5185 facet->nf_flow.output_iface = ctx.nf_output_iface;
5186 facet->has_learn = ctx.has_learn;
5187 facet->has_normal = ctx.has_normal;
5188 facet->has_fin_timeout = ctx.has_fin_timeout;
5189 facet->mirrors = ctx.mirrors;
5192 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
5193 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
5195 if (new_actions && new_actions[i].odp_actions) {
5196 free(subfacet->actions);
5197 subfacet->actions = new_actions[i].odp_actions;
5198 subfacet->actions_len = new_actions[i].actions_len;
5204 if (facet->rule != new_rule) {
5205 COVERAGE_INC(facet_changed_rule);
5206 list_remove(&facet->list_node);
5207 list_push_back(&new_rule->facets, &facet->list_node);
5208 facet->rule = new_rule;
5209 facet->used = new_rule->up.created;
5210 facet->prev_used = facet->used;
5214 /* Updates 'facet''s used time. Caller is responsible for calling
5215 * facet_push_stats() to update the flows which 'facet' resubmits into. */
5217 facet_update_time(struct facet *facet, long long int used)
5219 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5220 if (used > facet->used) {
5222 ofproto_rule_update_used(&facet->rule->up, used);
5223 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
5228 facet_reset_counters(struct facet *facet)
5230 facet->packet_count = 0;
5231 facet->byte_count = 0;
5232 facet->prev_packet_count = 0;
5233 facet->prev_byte_count = 0;
5234 facet->accounted_bytes = 0;
5238 facet_push_stats(struct facet *facet)
5240 struct dpif_flow_stats stats;
5242 ovs_assert(facet->packet_count >= facet->prev_packet_count);
5243 ovs_assert(facet->byte_count >= facet->prev_byte_count);
5244 ovs_assert(facet->used >= facet->prev_used);
5246 stats.n_packets = facet->packet_count - facet->prev_packet_count;
5247 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
5248 stats.used = facet->used;
5249 stats.tcp_flags = 0;
5251 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
5252 facet->prev_packet_count = facet->packet_count;
5253 facet->prev_byte_count = facet->byte_count;
5254 facet->prev_used = facet->used;
5256 flow_push_stats(facet, &stats);
5258 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
5259 facet->mirrors, stats.n_packets, stats.n_bytes);
5264 push_all_stats__(bool run_fast)
5266 static long long int rl = LLONG_MIN;
5267 struct ofproto_dpif *ofproto;
5269 if (time_msec() < rl) {
5273 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
5274 struct facet *facet;
5276 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
5277 facet_push_stats(facet);
5284 rl = time_msec() + 100;
5288 push_all_stats(void)
5290 push_all_stats__(true);
5294 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
5296 rule->packet_count += stats->n_packets;
5297 rule->byte_count += stats->n_bytes;
5298 ofproto_rule_update_used(&rule->up, stats->used);
5301 /* Pushes flow statistics to the rules which 'facet->flow' resubmits
5302 * into given 'facet->rule''s actions and mirrors. */
5304 flow_push_stats(struct facet *facet, const struct dpif_flow_stats *stats)
5306 struct rule_dpif *rule = facet->rule;
5307 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5308 struct subfacet *subfacet = facet_get_subfacet(facet);
5309 struct action_xlate_ctx ctx;
5311 ofproto_rule_update_used(&rule->up, stats->used);
5313 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
5314 &subfacet->initial_vals, rule, 0, NULL);
5315 ctx.resubmit_stats = stats;
5316 xlate_actions_for_side_effects(&ctx, rule->up.ofpacts,
5317 rule->up.ofpacts_len);
5322 static struct subfacet *
5323 subfacet_find(struct ofproto_dpif *ofproto,
5324 const struct nlattr *key, size_t key_len, uint32_t key_hash)
5326 struct subfacet *subfacet;
5328 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
5329 &ofproto->subfacets) {
5330 if (subfacet->key_len == key_len
5331 && !memcmp(key, subfacet->key, key_len)) {
5339 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
5340 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
5341 * existing subfacet if there is one, otherwise creates and returns a
5344 * If the returned subfacet is new, then subfacet->actions will be NULL, in
5345 * which case the caller must populate the actions with
5346 * subfacet_make_actions(). */
5347 static struct subfacet *
5348 subfacet_create(struct facet *facet, struct flow_miss *miss,
5351 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5352 enum odp_key_fitness key_fitness = miss->key_fitness;
5353 const struct nlattr *key = miss->key;
5354 size_t key_len = miss->key_len;
5356 struct subfacet *subfacet;
5358 key_hash = odp_flow_key_hash(key, key_len);
5360 if (list_is_empty(&facet->subfacets)) {
5361 subfacet = &facet->one_subfacet;
5363 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
5365 if (subfacet->facet == facet) {
5369 /* This shouldn't happen. */
5370 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
5371 subfacet_destroy(subfacet);
5374 subfacet = xmalloc(sizeof *subfacet);
5377 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
5378 list_push_back(&facet->subfacets, &subfacet->list_node);
5379 subfacet->facet = facet;
5380 subfacet->key_fitness = key_fitness;
5381 subfacet->key = xmemdup(key, key_len);
5382 subfacet->key_len = key_len;
5383 subfacet->used = now;
5384 subfacet->created = now;
5385 subfacet->dp_packet_count = 0;
5386 subfacet->dp_byte_count = 0;
5387 subfacet->actions_len = 0;
5388 subfacet->actions = NULL;
5389 subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE
5392 subfacet->path = SF_NOT_INSTALLED;
5393 subfacet->initial_vals = miss->initial_vals;
5394 subfacet->odp_in_port = miss->odp_in_port;
5396 ofproto->subfacet_add_count++;
5400 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
5401 * its facet within 'ofproto', and frees it. */
5403 subfacet_destroy__(struct subfacet *subfacet)
5405 struct facet *facet = subfacet->facet;
5406 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5408 /* Update ofproto stats before uninstall the subfacet. */
5409 ofproto->subfacet_del_count++;
5410 ofproto->total_subfacet_life_span += (time_msec() - subfacet->created);
5412 subfacet_uninstall(subfacet);
5413 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
5414 list_remove(&subfacet->list_node);
5415 free(subfacet->key);
5416 free(subfacet->actions);
5417 if (subfacet != &facet->one_subfacet) {
5422 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5423 * last remaining subfacet in its facet destroys the facet too. */
5425 subfacet_destroy(struct subfacet *subfacet)
5427 struct facet *facet = subfacet->facet;
5429 if (list_is_singleton(&facet->subfacets)) {
5430 /* facet_remove() needs at least one subfacet (it will remove it). */
5431 facet_remove(facet);
5433 subfacet_destroy__(subfacet);
5438 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
5439 struct subfacet **subfacets, int n)
5441 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5442 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5443 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5446 for (i = 0; i < n; i++) {
5447 ops[i].type = DPIF_OP_FLOW_DEL;
5448 ops[i].u.flow_del.key = subfacets[i]->key;
5449 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
5450 ops[i].u.flow_del.stats = &stats[i];
5454 dpif_operate(ofproto->backer->dpif, opsp, n);
5455 for (i = 0; i < n; i++) {
5456 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5457 subfacets[i]->path = SF_NOT_INSTALLED;
5458 subfacet_destroy(subfacets[i]);
5463 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
5464 * Translates the actions into 'odp_actions', which the caller must have
5465 * initialized and is responsible for uninitializing. */
5467 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
5468 struct ofpbuf *odp_actions)
5470 struct facet *facet = subfacet->facet;
5471 struct rule_dpif *rule = facet->rule;
5472 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5474 struct action_xlate_ctx ctx;
5476 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
5477 &subfacet->initial_vals, rule, 0, packet);
5478 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions);
5479 facet->tags = ctx.tags;
5480 facet->has_learn = ctx.has_learn;
5481 facet->has_normal = ctx.has_normal;
5482 facet->has_fin_timeout = ctx.has_fin_timeout;
5483 facet->nf_flow.output_iface = ctx.nf_output_iface;
5484 facet->mirrors = ctx.mirrors;
5486 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
5487 if (subfacet->actions_len != odp_actions->size
5488 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
5489 free(subfacet->actions);
5490 subfacet->actions_len = odp_actions->size;
5491 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
5495 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5496 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5497 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5498 * since 'subfacet' was last updated.
5500 * Returns 0 if successful, otherwise a positive errno value. */
5502 subfacet_install(struct subfacet *subfacet,
5503 const struct nlattr *actions, size_t actions_len,
5504 struct dpif_flow_stats *stats,
5505 enum slow_path_reason slow)
5507 struct facet *facet = subfacet->facet;
5508 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5509 enum subfacet_path path = subfacet_want_path(slow);
5510 uint64_t slow_path_stub[128 / 8];
5511 enum dpif_flow_put_flags flags;
5514 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5516 flags |= DPIF_FP_ZERO_STATS;
5519 if (path == SF_SLOW_PATH) {
5520 compose_slow_path(ofproto, &facet->flow, slow,
5521 slow_path_stub, sizeof slow_path_stub,
5522 &actions, &actions_len);
5525 ret = dpif_flow_put(ofproto->backer->dpif, flags, subfacet->key,
5526 subfacet->key_len, actions, actions_len, stats);
5529 subfacet_reset_dp_stats(subfacet, stats);
5533 subfacet->path = path;
5539 subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats)
5541 return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len,
5542 stats, subfacet->slow);
5545 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5547 subfacet_uninstall(struct subfacet *subfacet)
5549 if (subfacet->path != SF_NOT_INSTALLED) {
5550 struct rule_dpif *rule = subfacet->facet->rule;
5551 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5552 struct dpif_flow_stats stats;
5555 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5556 subfacet->key_len, &stats);
5557 subfacet_reset_dp_stats(subfacet, &stats);
5559 subfacet_update_stats(subfacet, &stats);
5561 subfacet->path = SF_NOT_INSTALLED;
5563 ovs_assert(subfacet->dp_packet_count == 0);
5564 ovs_assert(subfacet->dp_byte_count == 0);
5568 /* Resets 'subfacet''s datapath statistics counters. This should be called
5569 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5570 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5571 * was reset in the datapath. 'stats' will be modified to include only
5572 * statistics new since 'subfacet' was last updated. */
5574 subfacet_reset_dp_stats(struct subfacet *subfacet,
5575 struct dpif_flow_stats *stats)
5578 && subfacet->dp_packet_count <= stats->n_packets
5579 && subfacet->dp_byte_count <= stats->n_bytes) {
5580 stats->n_packets -= subfacet->dp_packet_count;
5581 stats->n_bytes -= subfacet->dp_byte_count;
5584 subfacet->dp_packet_count = 0;
5585 subfacet->dp_byte_count = 0;
5588 /* Updates 'subfacet''s used time. The caller is responsible for calling
5589 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
5591 subfacet_update_time(struct subfacet *subfacet, long long int used)
5593 if (used > subfacet->used) {
5594 subfacet->used = used;
5595 facet_update_time(subfacet->facet, used);
5599 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5601 * Because of the meaning of a subfacet's counters, it only makes sense to do
5602 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5603 * represents a packet that was sent by hand or if it represents statistics
5604 * that have been cleared out of the datapath. */
5606 subfacet_update_stats(struct subfacet *subfacet,
5607 const struct dpif_flow_stats *stats)
5609 if (stats->n_packets || stats->used > subfacet->used) {
5610 struct facet *facet = subfacet->facet;
5612 subfacet_update_time(subfacet, stats->used);
5613 facet->packet_count += stats->n_packets;
5614 facet->byte_count += stats->n_bytes;
5615 facet->tcp_flags |= stats->tcp_flags;
5616 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
5622 static struct rule_dpif *
5623 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
5625 struct rule_dpif *rule;
5627 rule = rule_dpif_lookup__(ofproto, flow, 0);
5632 return rule_dpif_miss_rule(ofproto, flow);
5635 static struct rule_dpif *
5636 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5639 struct cls_rule *cls_rule;
5640 struct classifier *cls;
5642 if (table_id >= N_TABLES) {
5646 cls = &ofproto->up.tables[table_id].cls;
5647 if (flow->nw_frag & FLOW_NW_FRAG_ANY
5648 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5649 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
5650 * are unavailable. */
5651 struct flow ofpc_normal_flow = *flow;
5652 ofpc_normal_flow.tp_src = htons(0);
5653 ofpc_normal_flow.tp_dst = htons(0);
5654 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
5656 cls_rule = classifier_lookup(cls, flow);
5658 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5661 static struct rule_dpif *
5662 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5664 struct ofport_dpif *port;
5666 port = get_ofp_port(ofproto, flow->in_port);
5668 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5669 return ofproto->miss_rule;
5672 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5673 return ofproto->no_packet_in_rule;
5675 return ofproto->miss_rule;
5679 complete_operation(struct rule_dpif *rule)
5681 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5683 rule_invalidate(rule);
5685 struct dpif_completion *c = xmalloc(sizeof *c);
5686 c->op = rule->up.pending;
5687 list_push_back(&ofproto->completions, &c->list_node);
5689 ofoperation_complete(rule->up.pending, 0);
5693 static struct rule *
5696 struct rule_dpif *rule = xmalloc(sizeof *rule);
5701 rule_dealloc(struct rule *rule_)
5703 struct rule_dpif *rule = rule_dpif_cast(rule_);
5708 rule_construct(struct rule *rule_)
5710 struct rule_dpif *rule = rule_dpif_cast(rule_);
5711 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5712 struct rule_dpif *victim;
5715 rule->packet_count = 0;
5716 rule->byte_count = 0;
5718 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5719 if (victim && !list_is_empty(&victim->facets)) {
5720 struct facet *facet;
5722 rule->facets = victim->facets;
5723 list_moved(&rule->facets);
5724 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5725 /* XXX: We're only clearing our local counters here. It's possible
5726 * that quite a few packets are unaccounted for in the datapath
5727 * statistics. These will be accounted to the new rule instead of
5728 * cleared as required. This could be fixed by clearing out the
5729 * datapath statistics for this facet, but currently it doesn't
5731 facet_reset_counters(facet);
5735 /* Must avoid list_moved() in this case. */
5736 list_init(&rule->facets);
5739 table_id = rule->up.table_id;
5741 rule->tag = victim->tag;
5742 } else if (table_id == 0) {
5747 miniflow_expand(&rule->up.cr.match.flow, &flow);
5748 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5749 ofproto->tables[table_id].basis);
5752 complete_operation(rule);
5757 rule_destruct(struct rule *rule_)
5759 struct rule_dpif *rule = rule_dpif_cast(rule_);
5760 struct facet *facet, *next_facet;
5762 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5763 facet_revalidate(facet);
5766 complete_operation(rule);
5770 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5772 struct rule_dpif *rule = rule_dpif_cast(rule_);
5773 struct facet *facet;
5775 /* push_all_stats() can handle flow misses which, when using the learn
5776 * action, can cause rules to be added and deleted. This can corrupt our
5777 * caller's datastructures which assume that rule_get_stats() doesn't have
5778 * an impact on the flow table. To be safe, we disable miss handling. */
5779 push_all_stats__(false);
5781 /* Start from historical data for 'rule' itself that are no longer tracked
5782 * in facets. This counts, for example, facets that have expired. */
5783 *packets = rule->packet_count;
5784 *bytes = rule->byte_count;
5786 /* Add any statistics that are tracked by facets. This includes
5787 * statistical data recently updated by ofproto_update_stats() as well as
5788 * stats for packets that were executed "by hand" via dpif_execute(). */
5789 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5790 *packets += facet->packet_count;
5791 *bytes += facet->byte_count;
5796 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5797 struct ofpbuf *packet)
5799 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5800 struct initial_vals initial_vals;
5801 struct dpif_flow_stats stats;
5802 struct action_xlate_ctx ctx;
5803 uint64_t odp_actions_stub[1024 / 8];
5804 struct ofpbuf odp_actions;
5806 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5807 rule_credit_stats(rule, &stats);
5809 initial_vals.vlan_tci = flow->vlan_tci;
5810 initial_vals.tunnel_ip_tos = flow->tunnel.ip_tos;
5811 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5812 action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals,
5813 rule, stats.tcp_flags, packet);
5814 ctx.resubmit_stats = &stats;
5815 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
5817 execute_odp_actions(ofproto, flow, odp_actions.data,
5818 odp_actions.size, packet);
5820 ofpbuf_uninit(&odp_actions);
5824 rule_execute(struct rule *rule, const struct flow *flow,
5825 struct ofpbuf *packet)
5827 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5828 ofpbuf_delete(packet);
5833 rule_modify_actions(struct rule *rule_)
5835 struct rule_dpif *rule = rule_dpif_cast(rule_);
5837 complete_operation(rule);
5840 /* Sends 'packet' out 'ofport'.
5841 * May modify 'packet'.
5842 * Returns 0 if successful, otherwise a positive errno value. */
5844 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5846 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5847 uint64_t odp_actions_stub[1024 / 8];
5848 struct ofpbuf key, odp_actions;
5849 struct odputil_keybuf keybuf;
5854 flow_extract(packet, 0, 0, NULL, OFPP_LOCAL, &flow);
5855 if (netdev_vport_is_patch(ofport->up.netdev)) {
5856 struct ofproto_dpif *peer_ofproto;
5857 struct dpif_flow_stats stats;
5858 struct ofport_dpif *peer;
5859 struct rule_dpif *rule;
5861 peer = ofport_get_peer(ofport);
5866 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5867 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5868 netdev_vport_inc_rx(peer->up.netdev, &stats);
5870 flow.in_port = peer->up.ofp_port;
5871 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5872 rule = rule_dpif_lookup(peer_ofproto, &flow);
5873 rule_dpif_execute(rule, &flow, packet);
5878 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5880 if (ofport->tnl_port) {
5881 struct dpif_flow_stats stats;
5883 odp_port = tnl_port_send(ofport->tnl_port, &flow);
5884 if (odp_port == OVSP_NONE) {
5888 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5889 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5890 odp_put_tunnel_action(&flow.tunnel, &odp_actions);
5891 odp_put_skb_mark_action(flow.skb_mark, &odp_actions);
5893 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
5895 if (odp_port != ofport->odp_port) {
5896 eth_pop_vlan(packet);
5897 flow.vlan_tci = htons(0);
5901 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5902 odp_flow_key_from_flow(&key, &flow,
5903 ofp_port_to_odp_port(ofproto, flow.in_port));
5905 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
5906 compose_ipfix_action(ofproto, &odp_actions, &flow);
5908 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
5909 error = dpif_execute(ofproto->backer->dpif,
5911 odp_actions.data, odp_actions.size,
5913 ofpbuf_uninit(&odp_actions);
5916 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
5917 ofproto->up.name, odp_port, strerror(error));
5919 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
5923 /* OpenFlow to datapath action translation. */
5925 static bool may_receive(const struct ofport_dpif *, struct action_xlate_ctx *);
5926 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5927 struct action_xlate_ctx *);
5928 static void xlate_normal(struct action_xlate_ctx *);
5930 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5931 * The action will state 'slow' as the reason that the action is in the slow
5932 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5933 * dump-flows" output to see why a flow is in the slow path.)
5935 * The 'stub_size' bytes in 'stub' will be used to store the action.
5936 * 'stub_size' must be large enough for the action.
5938 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5941 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5942 enum slow_path_reason slow,
5943 uint64_t *stub, size_t stub_size,
5944 const struct nlattr **actionsp, size_t *actions_lenp)
5946 union user_action_cookie cookie;
5949 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5950 cookie.slow_path.unused = 0;
5951 cookie.slow_path.reason = slow;
5953 ofpbuf_use_stack(&buf, stub, stub_size);
5954 if (slow & (SLOW_CFM | SLOW_BFD | SLOW_LACP | SLOW_STP)) {
5955 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5956 odp_put_userspace_action(pid, &cookie, sizeof cookie.slow_path, &buf);
5958 put_userspace_action(ofproto, &buf, flow, &cookie,
5959 sizeof cookie.slow_path);
5961 *actionsp = buf.data;
5962 *actions_lenp = buf.size;
5966 put_userspace_action(const struct ofproto_dpif *ofproto,
5967 struct ofpbuf *odp_actions,
5968 const struct flow *flow,
5969 const union user_action_cookie *cookie,
5970 const size_t cookie_size)
5974 pid = dpif_port_get_pid(ofproto->backer->dpif,
5975 ofp_port_to_odp_port(ofproto, flow->in_port));
5977 return odp_put_userspace_action(pid, cookie, cookie_size, odp_actions);
5980 /* Compose SAMPLE action for sFlow or IPFIX. The given probability is
5981 * the number of packets out of UINT32_MAX to sample. The given
5982 * cookie is passed back in the callback for each sampled packet.
5985 compose_sample_action(const struct ofproto_dpif *ofproto,
5986 struct ofpbuf *odp_actions,
5987 const struct flow *flow,
5988 const uint32_t probability,
5989 const union user_action_cookie *cookie,
5990 const size_t cookie_size)
5992 size_t sample_offset, actions_offset;
5995 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5997 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5999 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
6000 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, cookie,
6003 nl_msg_end_nested(odp_actions, actions_offset);
6004 nl_msg_end_nested(odp_actions, sample_offset);
6005 return cookie_offset;
6009 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
6010 ovs_be16 vlan_tci, uint32_t odp_port,
6011 unsigned int n_outputs, union user_action_cookie *cookie)
6015 cookie->type = USER_ACTION_COOKIE_SFLOW;
6016 cookie->sflow.vlan_tci = vlan_tci;
6018 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
6019 * port information") for the interpretation of cookie->output. */
6020 switch (n_outputs) {
6022 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
6023 cookie->sflow.output = 0x40000000 | 256;
6027 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
6029 cookie->sflow.output = ifindex;
6034 /* 0x80000000 means "multiple output ports. */
6035 cookie->sflow.output = 0x80000000 | n_outputs;
6040 /* Compose SAMPLE action for sFlow bridge sampling. */
6042 compose_sflow_action(const struct ofproto_dpif *ofproto,
6043 struct ofpbuf *odp_actions,
6044 const struct flow *flow,
6047 uint32_t probability;
6048 union user_action_cookie cookie;
6050 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
6054 probability = dpif_sflow_get_probability(ofproto->sflow);
6055 compose_sflow_cookie(ofproto, htons(0), odp_port,
6056 odp_port == OVSP_NONE ? 0 : 1, &cookie);
6058 return compose_sample_action(ofproto, odp_actions, flow, probability,
6059 &cookie, sizeof cookie.sflow);
6063 compose_flow_sample_cookie(uint16_t probability, uint32_t collector_set_id,
6064 uint32_t obs_domain_id, uint32_t obs_point_id,
6065 union user_action_cookie *cookie)
6067 cookie->type = USER_ACTION_COOKIE_FLOW_SAMPLE;
6068 cookie->flow_sample.probability = probability;
6069 cookie->flow_sample.collector_set_id = collector_set_id;
6070 cookie->flow_sample.obs_domain_id = obs_domain_id;
6071 cookie->flow_sample.obs_point_id = obs_point_id;
6075 compose_ipfix_cookie(union user_action_cookie *cookie)
6077 cookie->type = USER_ACTION_COOKIE_IPFIX;
6080 /* Compose SAMPLE action for IPFIX bridge sampling. */
6082 compose_ipfix_action(const struct ofproto_dpif *ofproto,
6083 struct ofpbuf *odp_actions,
6084 const struct flow *flow)
6086 uint32_t probability;
6087 union user_action_cookie cookie;
6089 if (!ofproto->ipfix || flow->in_port == OFPP_NONE) {
6093 probability = dpif_ipfix_get_bridge_exporter_probability(ofproto->ipfix);
6094 compose_ipfix_cookie(&cookie);
6096 compose_sample_action(ofproto, odp_actions, flow, probability,
6097 &cookie, sizeof cookie.ipfix);
6100 /* SAMPLE action for sFlow must be first action in any given list of
6101 * actions. At this point we do not have all information required to
6102 * build it. So try to build sample action as complete as possible. */
6104 add_sflow_action(struct action_xlate_ctx *ctx)
6106 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
6108 &ctx->flow, OVSP_NONE);
6109 ctx->sflow_odp_port = 0;
6110 ctx->sflow_n_outputs = 0;
6113 /* SAMPLE action for IPFIX must be 1st or 2nd action in any given list
6114 * of actions, eventually after the SAMPLE action for sFlow. */
6116 add_ipfix_action(struct action_xlate_ctx *ctx)
6118 compose_ipfix_action(ctx->ofproto, ctx->odp_actions, &ctx->flow);
6121 /* Fix SAMPLE action according to data collected while composing ODP actions.
6122 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
6123 * USERSPACE action's user-cookie which is required for sflow. */
6125 fix_sflow_action(struct action_xlate_ctx *ctx)
6127 const struct flow *base = &ctx->base_flow;
6128 union user_action_cookie *cookie;
6130 if (!ctx->user_cookie_offset) {
6134 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
6135 sizeof cookie->sflow);
6136 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
6138 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
6139 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
6143 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
6146 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
6147 ovs_be16 flow_vlan_tci;
6148 uint32_t flow_skb_mark;
6149 uint8_t flow_nw_tos;
6150 struct priority_to_dscp *pdscp;
6151 uint32_t out_port, odp_port;
6153 /* If 'struct flow' gets additional metadata, we'll need to zero it out
6154 * before traversing a patch port. */
6155 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 20);
6158 xlate_report(ctx, "Nonexistent output port");
6160 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
6161 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
6163 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
6164 xlate_report(ctx, "STP not in forwarding state, skipping output");
6168 if (netdev_vport_is_patch(ofport->up.netdev)) {
6169 struct ofport_dpif *peer = ofport_get_peer(ofport);
6170 struct flow old_flow = ctx->flow;
6171 const struct ofproto_dpif *peer_ofproto;
6172 enum slow_path_reason special;
6173 struct ofport_dpif *in_port;
6176 xlate_report(ctx, "Nonexistent patch port peer");
6180 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
6181 if (peer_ofproto->backer != ctx->ofproto->backer) {
6182 xlate_report(ctx, "Patch port peer on a different datapath");
6186 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
6187 ctx->flow.in_port = peer->up.ofp_port;
6188 ctx->flow.metadata = htonll(0);
6189 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
6190 memset(ctx->flow.regs, 0, sizeof ctx->flow.regs);
6192 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
6193 special = process_special(ctx->ofproto, &ctx->flow, in_port,
6196 ctx->slow |= special;
6197 } else if (!in_port || may_receive(in_port, ctx)) {
6198 if (!in_port || stp_forward_in_state(in_port->stp_state)) {
6199 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
6201 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
6202 * learning action look at the packet, then drop it. */
6203 struct flow old_base_flow = ctx->base_flow;
6204 size_t old_size = ctx->odp_actions->size;
6205 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
6206 ctx->base_flow = old_base_flow;
6207 ctx->odp_actions->size = old_size;
6211 ctx->flow = old_flow;
6212 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
6214 if (ctx->resubmit_stats) {
6215 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
6216 netdev_vport_inc_rx(peer->up.netdev, ctx->resubmit_stats);
6222 flow_vlan_tci = ctx->flow.vlan_tci;
6223 flow_skb_mark = ctx->flow.skb_mark;
6224 flow_nw_tos = ctx->flow.nw_tos;
6226 pdscp = get_priority(ofport, ctx->flow.skb_priority);
6228 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
6229 ctx->flow.nw_tos |= pdscp->dscp;
6232 if (ofport->tnl_port) {
6233 /* Save tunnel metadata so that changes made due to
6234 * the Logical (tunnel) Port are not visible for any further
6235 * matches, while explicit set actions on tunnel metadata are.
6237 struct flow_tnl flow_tnl = ctx->flow.tunnel;
6238 odp_port = tnl_port_send(ofport->tnl_port, &ctx->flow);
6239 if (odp_port == OVSP_NONE) {
6240 xlate_report(ctx, "Tunneling decided against output");
6241 goto out; /* restore flow_nw_tos */
6244 if (ctx->resubmit_stats) {
6245 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
6247 out_port = odp_port;
6248 commit_odp_tunnel_action(&ctx->flow, &ctx->base_flow,
6250 ctx->flow.tunnel = flow_tnl; /* Restore tunnel metadata */
6252 odp_port = ofport->odp_port;
6253 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
6254 ctx->flow.vlan_tci);
6255 if (out_port != odp_port) {
6256 ctx->flow.vlan_tci = htons(0);
6258 ctx->flow.skb_mark &= ~IPSEC_MARK;
6260 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
6261 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
6263 ctx->sflow_odp_port = odp_port;
6264 ctx->sflow_n_outputs++;
6265 ctx->nf_output_iface = ofp_port;
6268 ctx->flow.vlan_tci = flow_vlan_tci;
6269 ctx->flow.skb_mark = flow_skb_mark;
6271 ctx->flow.nw_tos = flow_nw_tos;
6275 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
6277 compose_output_action__(ctx, ofp_port, true);
6281 tag_the_flow(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
6283 struct ofproto_dpif *ofproto = ctx->ofproto;
6284 uint8_t table_id = ctx->table_id;
6286 if (table_id > 0 && table_id < N_TABLES) {
6287 struct table_dpif *table = &ofproto->tables[table_id];
6288 if (table->other_table) {
6289 ctx->tags |= (rule && rule->tag
6291 : rule_calculate_tag(&ctx->flow,
6292 &table->other_table->mask,
6298 /* Common rule processing in one place to avoid duplicating code. */
6299 static struct rule_dpif *
6300 ctx_rule_hooks(struct action_xlate_ctx *ctx, struct rule_dpif *rule,
6303 if (ctx->resubmit_hook) {
6304 ctx->resubmit_hook(ctx, rule);
6306 if (rule == NULL && may_packet_in) {
6308 * check if table configuration flags
6309 * OFPTC_TABLE_MISS_CONTROLLER, default.
6310 * OFPTC_TABLE_MISS_CONTINUE,
6311 * OFPTC_TABLE_MISS_DROP
6312 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
6314 rule = rule_dpif_miss_rule(ctx->ofproto, &ctx->flow);
6316 if (rule && ctx->resubmit_stats) {
6317 rule_credit_stats(rule, ctx->resubmit_stats);
6323 xlate_table_action(struct action_xlate_ctx *ctx,
6324 uint16_t in_port, uint8_t table_id, bool may_packet_in)
6326 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
6327 struct rule_dpif *rule;
6328 uint16_t old_in_port = ctx->flow.in_port;
6329 uint8_t old_table_id = ctx->table_id;
6331 ctx->table_id = table_id;
6333 /* Look up a flow with 'in_port' as the input port. */
6334 ctx->flow.in_port = in_port;
6335 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->flow, table_id);
6337 tag_the_flow(ctx, rule);
6339 /* Restore the original input port. Otherwise OFPP_NORMAL and
6340 * OFPP_IN_PORT will have surprising behavior. */
6341 ctx->flow.in_port = old_in_port;
6343 rule = ctx_rule_hooks(ctx, rule, may_packet_in);
6346 struct rule_dpif *old_rule = ctx->rule;
6350 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
6351 ctx->rule = old_rule;
6355 ctx->table_id = old_table_id;
6357 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6359 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
6360 MAX_RESUBMIT_RECURSION);
6361 ctx->max_resubmit_trigger = true;
6366 xlate_ofpact_resubmit(struct action_xlate_ctx *ctx,
6367 const struct ofpact_resubmit *resubmit)
6372 in_port = resubmit->in_port;
6373 if (in_port == OFPP_IN_PORT) {
6374 in_port = ctx->flow.in_port;
6377 table_id = resubmit->table_id;
6378 if (table_id == 255) {
6379 table_id = ctx->table_id;
6382 xlate_table_action(ctx, in_port, table_id, false);
6386 flood_packets(struct action_xlate_ctx *ctx, bool all)
6388 struct ofport_dpif *ofport;
6390 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
6391 uint16_t ofp_port = ofport->up.ofp_port;
6393 if (ofp_port == ctx->flow.in_port) {
6398 compose_output_action__(ctx, ofp_port, false);
6399 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
6400 compose_output_action(ctx, ofp_port);
6404 ctx->nf_output_iface = NF_OUT_FLOOD;
6408 execute_controller_action(struct action_xlate_ctx *ctx, int len,
6409 enum ofp_packet_in_reason reason,
6410 uint16_t controller_id)
6412 struct ofputil_packet_in pin;
6413 struct ofpbuf *packet;
6415 ctx->slow |= SLOW_CONTROLLER;
6420 packet = ofpbuf_clone(ctx->packet);
6422 if (packet->l2 && packet->l3) {
6423 struct eth_header *eh;
6424 uint16_t mpls_depth;
6426 eth_pop_vlan(packet);
6429 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
6430 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
6432 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
6433 eth_push_vlan(packet, ctx->flow.vlan_tci);
6436 mpls_depth = eth_mpls_depth(packet);
6438 if (mpls_depth < ctx->flow.mpls_depth) {
6439 push_mpls(packet, ctx->flow.dl_type, ctx->flow.mpls_lse);
6440 } else if (mpls_depth > ctx->flow.mpls_depth) {
6441 pop_mpls(packet, ctx->flow.dl_type);
6442 } else if (mpls_depth) {
6443 set_mpls_lse(packet, ctx->flow.mpls_lse);
6447 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6448 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
6449 ctx->flow.nw_tos, ctx->flow.nw_ttl);
6453 if (ctx->flow.nw_proto == IPPROTO_TCP) {
6454 packet_set_tcp_port(packet, ctx->flow.tp_src,
6456 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
6457 packet_set_udp_port(packet, ctx->flow.tp_src,
6464 pin.packet = packet->data;
6465 pin.packet_len = packet->size;
6466 pin.reason = reason;
6467 pin.controller_id = controller_id;
6468 pin.table_id = ctx->table_id;
6469 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
6472 flow_get_metadata(&ctx->flow, &pin.fmd);
6474 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
6475 ofpbuf_delete(packet);
6479 execute_mpls_push_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type)
6481 ovs_assert(eth_type_mpls(eth_type));
6483 if (ctx->base_flow.mpls_depth) {
6484 ctx->flow.mpls_lse &= ~htonl(MPLS_BOS_MASK);
6485 ctx->flow.mpls_depth++;
6490 if (ctx->flow.dl_type == htons(ETH_TYPE_IPV6)) {
6491 label = htonl(0x2); /* IPV6 Explicit Null. */
6493 label = htonl(0x0); /* IPV4 Explicit Null. */
6495 tc = (ctx->flow.nw_tos & IP_DSCP_MASK) >> 2;
6496 ttl = ctx->flow.nw_ttl ? ctx->flow.nw_ttl : 0x40;
6497 ctx->flow.mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
6498 ctx->flow.mpls_depth = 1;
6500 ctx->flow.dl_type = eth_type;
6504 execute_mpls_pop_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type)
6506 ovs_assert(eth_type_mpls(ctx->flow.dl_type));
6507 ovs_assert(!eth_type_mpls(eth_type));
6509 if (ctx->flow.mpls_depth) {
6510 ctx->flow.mpls_depth--;
6511 ctx->flow.mpls_lse = htonl(0);
6512 if (!ctx->flow.mpls_depth) {
6513 ctx->flow.dl_type = eth_type;
6519 compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
6521 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
6522 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
6526 if (ctx->flow.nw_ttl > 1) {
6532 for (i = 0; i < ids->n_controllers; i++) {
6533 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
6537 /* Stop processing for current table. */
6543 execute_set_mpls_ttl_action(struct action_xlate_ctx *ctx, uint8_t ttl)
6545 if (!eth_type_mpls(ctx->flow.dl_type)) {
6549 set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl);
6554 execute_dec_mpls_ttl_action(struct action_xlate_ctx *ctx)
6556 uint8_t ttl = mpls_lse_to_ttl(ctx->flow.mpls_lse);
6558 if (!eth_type_mpls(ctx->flow.dl_type)) {
6564 set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl);
6567 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
6569 /* Stop processing for current table. */
6575 xlate_output_action(struct action_xlate_ctx *ctx,
6576 uint16_t port, uint16_t max_len, bool may_packet_in)
6578 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
6580 ctx->nf_output_iface = NF_OUT_DROP;
6584 compose_output_action(ctx, ctx->flow.in_port);
6587 xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in);
6593 flood_packets(ctx, false);
6596 flood_packets(ctx, true);
6598 case OFPP_CONTROLLER:
6599 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
6605 if (port != ctx->flow.in_port) {
6606 compose_output_action(ctx, port);
6608 xlate_report(ctx, "skipping output to input port");
6613 if (prev_nf_output_iface == NF_OUT_FLOOD) {
6614 ctx->nf_output_iface = NF_OUT_FLOOD;
6615 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
6616 ctx->nf_output_iface = prev_nf_output_iface;
6617 } else if (prev_nf_output_iface != NF_OUT_DROP &&
6618 ctx->nf_output_iface != NF_OUT_FLOOD) {
6619 ctx->nf_output_iface = NF_OUT_MULTI;
6624 xlate_output_reg_action(struct action_xlate_ctx *ctx,
6625 const struct ofpact_output_reg *or)
6627 uint64_t port = mf_get_subfield(&or->src, &ctx->flow);
6628 if (port <= UINT16_MAX) {
6629 xlate_output_action(ctx, port, or->max_len, false);
6634 xlate_enqueue_action(struct action_xlate_ctx *ctx,
6635 const struct ofpact_enqueue *enqueue)
6637 uint16_t ofp_port = enqueue->port;
6638 uint32_t queue_id = enqueue->queue;
6639 uint32_t flow_priority, priority;
6642 /* Translate queue to priority. */
6643 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6644 queue_id, &priority);
6646 /* Fall back to ordinary output action. */
6647 xlate_output_action(ctx, enqueue->port, 0, false);
6651 /* Check output port. */
6652 if (ofp_port == OFPP_IN_PORT) {
6653 ofp_port = ctx->flow.in_port;
6654 } else if (ofp_port == ctx->flow.in_port) {
6658 /* Add datapath actions. */
6659 flow_priority = ctx->flow.skb_priority;
6660 ctx->flow.skb_priority = priority;
6661 compose_output_action(ctx, ofp_port);
6662 ctx->flow.skb_priority = flow_priority;
6664 /* Update NetFlow output port. */
6665 if (ctx->nf_output_iface == NF_OUT_DROP) {
6666 ctx->nf_output_iface = ofp_port;
6667 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
6668 ctx->nf_output_iface = NF_OUT_MULTI;
6673 xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id)
6675 uint32_t skb_priority;
6677 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6678 queue_id, &skb_priority)) {
6679 ctx->flow.skb_priority = skb_priority;
6681 /* Couldn't translate queue to a priority. Nothing to do. A warning
6682 * has already been logged. */
6687 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
6689 struct ofproto_dpif *ofproto = ofproto_;
6690 struct ofport_dpif *port;
6700 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
6703 port = get_ofp_port(ofproto, ofp_port);
6704 return port ? port->may_enable : false;
6709 xlate_bundle_action(struct action_xlate_ctx *ctx,
6710 const struct ofpact_bundle *bundle)
6714 port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto);
6715 if (bundle->dst.field) {
6716 nxm_reg_load(&bundle->dst, port, &ctx->flow);
6718 xlate_output_action(ctx, port, 0, false);
6723 xlate_learn_action(struct action_xlate_ctx *ctx,
6724 const struct ofpact_learn *learn)
6726 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
6727 struct ofputil_flow_mod fm;
6728 uint64_t ofpacts_stub[1024 / 8];
6729 struct ofpbuf ofpacts;
6732 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
6733 learn_execute(learn, &ctx->flow, &fm, &ofpacts);
6735 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
6736 if (error && !VLOG_DROP_WARN(&rl)) {
6737 VLOG_WARN("learning action failed to modify flow table (%s)",
6738 ofperr_get_name(error));
6741 ofpbuf_uninit(&ofpacts);
6744 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
6745 * means "infinite". */
6747 reduce_timeout(uint16_t max, uint16_t *timeout)
6749 if (max && (!*timeout || *timeout > max)) {
6755 xlate_fin_timeout(struct action_xlate_ctx *ctx,
6756 const struct ofpact_fin_timeout *oft)
6758 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
6759 struct rule_dpif *rule = ctx->rule;
6761 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
6762 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
6767 xlate_sample_action(struct action_xlate_ctx *ctx,
6768 const struct ofpact_sample *os)
6770 union user_action_cookie cookie;
6771 /* Scale the probability from 16-bit to 32-bit while representing
6772 * the same percentage. */
6773 uint32_t probability = (os->probability << 16) | os->probability;
6775 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
6777 compose_flow_sample_cookie(os->probability, os->collector_set_id,
6778 os->obs_domain_id, os->obs_point_id, &cookie);
6779 compose_sample_action(ctx->ofproto, ctx->odp_actions, &ctx->flow,
6780 probability, &cookie, sizeof cookie.flow_sample);
6784 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
6786 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
6787 ? OFPUTIL_PC_NO_RECV_STP
6788 : OFPUTIL_PC_NO_RECV)) {
6792 /* Only drop packets here if both forwarding and learning are
6793 * disabled. If just learning is enabled, we need to have
6794 * OFPP_NORMAL and the learning action have a look at the packet
6795 * before we can drop it. */
6796 if (!stp_forward_in_state(port->stp_state)
6797 && !stp_learn_in_state(port->stp_state)) {
6805 tunnel_ecn_ok(struct action_xlate_ctx *ctx)
6807 if (is_ip_any(&ctx->base_flow)
6808 && (ctx->base_flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
6809 if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
6810 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
6811 " but is not ECN capable");
6814 /* Set the ECN CE value in the tunneled packet. */
6815 ctx->flow.nw_tos |= IP_ECN_CE;
6823 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
6824 struct action_xlate_ctx *ctx)
6826 bool was_evictable = true;
6827 const struct ofpact *a;
6830 /* Don't let the rule we're working on get evicted underneath us. */
6831 was_evictable = ctx->rule->up.evictable;
6832 ctx->rule->up.evictable = false;
6835 do_xlate_actions_again:
6836 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
6837 struct ofpact_controller *controller;
6838 const struct ofpact_metadata *metadata;
6846 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
6847 ofpact_get_OUTPUT(a)->max_len, true);
6850 case OFPACT_CONTROLLER:
6851 controller = ofpact_get_CONTROLLER(a);
6852 execute_controller_action(ctx, controller->max_len,
6854 controller->controller_id);
6857 case OFPACT_ENQUEUE:
6858 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
6861 case OFPACT_SET_VLAN_VID:
6862 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
6863 ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
6867 case OFPACT_SET_VLAN_PCP:
6868 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
6869 ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
6874 case OFPACT_STRIP_VLAN:
6875 ctx->flow.vlan_tci = htons(0);
6878 case OFPACT_PUSH_VLAN:
6879 /* XXX 802.1AD(QinQ) */
6880 ctx->flow.vlan_tci = htons(VLAN_CFI);
6883 case OFPACT_SET_ETH_SRC:
6884 memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
6888 case OFPACT_SET_ETH_DST:
6889 memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
6893 case OFPACT_SET_IPV4_SRC:
6894 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6895 ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
6899 case OFPACT_SET_IPV4_DST:
6900 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6901 ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
6905 case OFPACT_SET_IPV4_DSCP:
6906 /* OpenFlow 1.0 only supports IPv4. */
6907 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6908 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
6909 ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
6913 case OFPACT_SET_L4_SRC_PORT:
6914 if (is_ip_any(&ctx->flow)) {
6915 ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
6919 case OFPACT_SET_L4_DST_PORT:
6920 if (is_ip_any(&ctx->flow)) {
6921 ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
6925 case OFPACT_RESUBMIT:
6926 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
6929 case OFPACT_SET_TUNNEL:
6930 ctx->flow.tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
6933 case OFPACT_SET_QUEUE:
6934 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
6937 case OFPACT_POP_QUEUE:
6938 ctx->flow.skb_priority = ctx->orig_skb_priority;
6941 case OFPACT_REG_MOVE:
6942 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow);
6945 case OFPACT_REG_LOAD:
6946 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow);
6949 case OFPACT_STACK_PUSH:
6950 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), &ctx->flow,
6954 case OFPACT_STACK_POP:
6955 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), &ctx->flow,
6959 case OFPACT_PUSH_MPLS:
6960 execute_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
6963 case OFPACT_POP_MPLS:
6964 execute_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
6967 case OFPACT_SET_MPLS_TTL:
6968 if (execute_set_mpls_ttl_action(ctx, ofpact_get_SET_MPLS_TTL(a)->ttl)) {
6973 case OFPACT_DEC_MPLS_TTL:
6974 if (execute_dec_mpls_ttl_action(ctx)) {
6979 case OFPACT_DEC_TTL:
6980 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6986 /* Nothing to do. */
6989 case OFPACT_MULTIPATH:
6990 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow);
6994 ctx->ofproto->has_bundle_action = true;
6995 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6998 case OFPACT_OUTPUT_REG:
6999 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
7003 ctx->has_learn = true;
7004 if (ctx->may_learn) {
7005 xlate_learn_action(ctx, ofpact_get_LEARN(a));
7013 case OFPACT_FIN_TIMEOUT:
7014 ctx->has_fin_timeout = true;
7015 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
7018 case OFPACT_CLEAR_ACTIONS:
7020 * Nothing to do because writa-actions is not supported for now.
7021 * When writa-actions is supported, clear-actions also must
7022 * be supported at the same time.
7026 case OFPACT_WRITE_METADATA:
7027 metadata = ofpact_get_WRITE_METADATA(a);
7028 ctx->flow.metadata &= ~metadata->mask;
7029 ctx->flow.metadata |= metadata->metadata & metadata->mask;
7032 case OFPACT_GOTO_TABLE: {
7033 /* It is assumed that goto-table is the last action. */
7034 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
7035 struct rule_dpif *rule;
7037 ovs_assert(ctx->table_id < ogt->table_id);
7039 ctx->table_id = ogt->table_id;
7041 /* Look up a flow from the new table. */
7042 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->flow, ctx->table_id);
7044 tag_the_flow(ctx, rule);
7046 rule = ctx_rule_hooks(ctx, rule, true);
7050 ctx->rule->up.evictable = was_evictable;
7053 was_evictable = rule->up.evictable;
7054 rule->up.evictable = false;
7056 /* Tail recursion removal. */
7057 ofpacts = rule->up.ofpacts;
7058 ofpacts_len = rule->up.ofpacts_len;
7059 goto do_xlate_actions_again;
7065 xlate_sample_action(ctx, ofpact_get_SAMPLE(a));
7072 ctx->rule->up.evictable = was_evictable;
7077 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
7078 struct ofproto_dpif *ofproto, const struct flow *flow,
7079 const struct initial_vals *initial_vals,
7080 struct rule_dpif *rule,
7081 uint8_t tcp_flags, const struct ofpbuf *packet)
7083 ovs_be64 initial_tun_id = flow->tunnel.tun_id;
7085 /* Flow initialization rules:
7086 * - 'base_flow' must match the kernel's view of the packet at the
7087 * time that action processing starts. 'flow' represents any
7088 * transformations we wish to make through actions.
7089 * - By default 'base_flow' and 'flow' are the same since the input
7090 * packet matches the output before any actions are applied.
7091 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
7092 * of the received packet as seen by the kernel. If we later output
7093 * to another device without any modifications this will cause us to
7094 * insert a new tag since the original one was stripped off by the
7096 * - Tunnel 'flow' is largely cleared when transitioning between
7097 * the input and output stages since it does not make sense to output
7098 * a packet with the exact headers that it was received with (i.e.
7099 * the destination IP is us). The one exception is the tun_id, which
7100 * is preserved to allow use in later resubmit lookups and loads into
7102 * - Tunnel 'base_flow' is completely cleared since that is what the
7103 * kernel does. If we wish to maintain the original values an action
7104 * needs to be generated. */
7106 ctx->ofproto = ofproto;
7108 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
7109 ctx->base_flow = ctx->flow;
7110 ctx->base_flow.vlan_tci = initial_vals->vlan_tci;
7111 ctx->base_flow.tunnel.ip_tos = initial_vals->tunnel_ip_tos;
7112 ctx->flow.tunnel.tun_id = initial_tun_id;
7114 ctx->packet = packet;
7115 ctx->may_learn = packet != NULL;
7116 ctx->tcp_flags = tcp_flags;
7117 ctx->resubmit_hook = NULL;
7118 ctx->report_hook = NULL;
7119 ctx->resubmit_stats = NULL;
7122 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
7123 * into datapath actions in 'odp_actions', using 'ctx'. */
7125 xlate_actions(struct action_xlate_ctx *ctx,
7126 const struct ofpact *ofpacts, size_t ofpacts_len,
7127 struct ofpbuf *odp_actions)
7129 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
7130 * that in the future we always keep a copy of the original flow for
7131 * tracing purposes. */
7132 static bool hit_resubmit_limit;
7134 enum slow_path_reason special;
7135 struct ofport_dpif *in_port;
7136 struct flow orig_flow;
7138 COVERAGE_INC(ofproto_dpif_xlate);
7140 ofpbuf_clear(odp_actions);
7141 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
7143 ctx->odp_actions = odp_actions;
7146 ctx->has_learn = false;
7147 ctx->has_normal = false;
7148 ctx->has_fin_timeout = false;
7149 ctx->nf_output_iface = NF_OUT_DROP;
7152 ctx->max_resubmit_trigger = false;
7153 ctx->orig_skb_priority = ctx->flow.skb_priority;
7157 ofpbuf_use_stub(&ctx->stack, ctx->init_stack, sizeof ctx->init_stack);
7159 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
7160 /* Do this conditionally because the copy is expensive enough that it
7161 * shows up in profiles. */
7162 orig_flow = ctx->flow;
7165 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
7166 switch (ctx->ofproto->up.frag_handling) {
7167 case OFPC_FRAG_NORMAL:
7168 /* We must pretend that transport ports are unavailable. */
7169 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
7170 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
7173 case OFPC_FRAG_DROP:
7176 case OFPC_FRAG_REASM:
7179 case OFPC_FRAG_NX_MATCH:
7180 /* Nothing to do. */
7183 case OFPC_INVALID_TTL_TO_CONTROLLER:
7188 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
7189 special = process_special(ctx->ofproto, &ctx->flow, in_port, ctx->packet);
7191 ctx->slow |= special;
7193 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
7194 struct initial_vals initial_vals;
7195 size_t sample_actions_len;
7196 uint32_t local_odp_port;
7198 initial_vals.vlan_tci = ctx->base_flow.vlan_tci;
7199 initial_vals.tunnel_ip_tos = ctx->base_flow.tunnel.ip_tos;
7201 add_sflow_action(ctx);
7202 add_ipfix_action(ctx);
7203 sample_actions_len = ctx->odp_actions->size;
7205 if (tunnel_ecn_ok(ctx) && (!in_port || may_receive(in_port, ctx))) {
7206 do_xlate_actions(ofpacts, ofpacts_len, ctx);
7208 /* We've let OFPP_NORMAL and the learning action look at the
7209 * packet, so drop it now if forwarding is disabled. */
7210 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
7211 ctx->odp_actions->size = sample_actions_len;
7215 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
7216 if (!hit_resubmit_limit) {
7217 /* We didn't record the original flow. Make sure we do from
7219 hit_resubmit_limit = true;
7220 } else if (!VLOG_DROP_ERR(&trace_rl)) {
7221 struct ds ds = DS_EMPTY_INITIALIZER;
7223 ofproto_trace(ctx->ofproto, &orig_flow, ctx->packet,
7224 &initial_vals, &ds);
7225 VLOG_ERR("Trace triggered by excessive resubmit "
7226 "recursion:\n%s", ds_cstr(&ds));
7231 local_odp_port = ofp_port_to_odp_port(ctx->ofproto, OFPP_LOCAL);
7232 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
7234 ctx->odp_actions->data,
7235 ctx->odp_actions->size)) {
7236 ctx->slow |= SLOW_IN_BAND;
7238 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
7240 compose_output_action(ctx, OFPP_LOCAL);
7243 if (ctx->ofproto->has_mirrors) {
7244 add_mirror_actions(ctx, &orig_flow);
7246 fix_sflow_action(ctx);
7249 ofpbuf_uninit(&ctx->stack);
7252 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
7253 * into datapath actions, using 'ctx', and discards the datapath actions. */
7255 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
7256 const struct ofpact *ofpacts,
7259 uint64_t odp_actions_stub[1024 / 8];
7260 struct ofpbuf odp_actions;
7262 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
7263 xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions);
7264 ofpbuf_uninit(&odp_actions);
7268 xlate_report(struct action_xlate_ctx *ctx, const char *s)
7270 if (ctx->report_hook) {
7271 ctx->report_hook(ctx, s);
7275 /* OFPP_NORMAL implementation. */
7277 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
7279 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
7280 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
7281 * the bundle on which the packet was received, returns the VLAN to which the
7284 * Both 'vid' and the return value are in the range 0...4095. */
7286 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
7288 switch (in_bundle->vlan_mode) {
7289 case PORT_VLAN_ACCESS:
7290 return in_bundle->vlan;
7293 case PORT_VLAN_TRUNK:
7296 case PORT_VLAN_NATIVE_UNTAGGED:
7297 case PORT_VLAN_NATIVE_TAGGED:
7298 return vid ? vid : in_bundle->vlan;
7305 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
7306 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
7309 * 'vid' should be the VID obtained from the 802.1Q header that was received as
7310 * part of a packet (specify 0 if there was no 802.1Q header), in the range
7313 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
7315 /* Allow any VID on the OFPP_NONE port. */
7316 if (in_bundle == &ofpp_none_bundle) {
7320 switch (in_bundle->vlan_mode) {
7321 case PORT_VLAN_ACCESS:
7324 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7325 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
7326 "packet received on port %s configured as VLAN "
7327 "%"PRIu16" access port",
7328 in_bundle->ofproto->up.name, vid,
7329 in_bundle->name, in_bundle->vlan);
7335 case PORT_VLAN_NATIVE_UNTAGGED:
7336 case PORT_VLAN_NATIVE_TAGGED:
7338 /* Port must always carry its native VLAN. */
7342 case PORT_VLAN_TRUNK:
7343 if (!ofbundle_includes_vlan(in_bundle, vid)) {
7345 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7346 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
7347 "received on port %s not configured for trunking "
7349 in_bundle->ofproto->up.name, vid,
7350 in_bundle->name, vid);
7362 /* Given 'vlan', the VLAN that a packet belongs to, and
7363 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
7364 * that should be included in the 802.1Q header. (If the return value is 0,
7365 * then the 802.1Q header should only be included in the packet if there is a
7368 * Both 'vlan' and the return value are in the range 0...4095. */
7370 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
7372 switch (out_bundle->vlan_mode) {
7373 case PORT_VLAN_ACCESS:
7376 case PORT_VLAN_TRUNK:
7377 case PORT_VLAN_NATIVE_TAGGED:
7380 case PORT_VLAN_NATIVE_UNTAGGED:
7381 return vlan == out_bundle->vlan ? 0 : vlan;
7389 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
7392 struct ofport_dpif *port;
7394 ovs_be16 tci, old_tci;
7396 vid = output_vlan_to_vid(out_bundle, vlan);
7397 if (!out_bundle->bond) {
7398 port = ofbundle_get_a_port(out_bundle);
7400 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
7403 /* No slaves enabled, so drop packet. */
7408 old_tci = ctx->flow.vlan_tci;
7410 if (tci || out_bundle->use_priority_tags) {
7411 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
7413 tci |= htons(VLAN_CFI);
7416 ctx->flow.vlan_tci = tci;
7418 compose_output_action(ctx, port->up.ofp_port);
7419 ctx->flow.vlan_tci = old_tci;
7423 mirror_mask_ffs(mirror_mask_t mask)
7425 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
7430 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
7432 return (bundle->vlan_mode != PORT_VLAN_ACCESS
7433 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
7437 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
7439 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
7442 /* Returns an arbitrary interface within 'bundle'. */
7443 static struct ofport_dpif *
7444 ofbundle_get_a_port(const struct ofbundle *bundle)
7446 return CONTAINER_OF(list_front(&bundle->ports),
7447 struct ofport_dpif, bundle_node);
7451 vlan_is_mirrored(const struct ofmirror *m, int vlan)
7453 return !m->vlans || bitmap_is_set(m->vlans, vlan);
7457 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
7459 struct ofproto_dpif *ofproto = ctx->ofproto;
7460 mirror_mask_t mirrors;
7461 struct ofbundle *in_bundle;
7464 const struct nlattr *a;
7467 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
7468 ctx->packet != NULL, NULL);
7472 mirrors = in_bundle->src_mirrors;
7474 /* Drop frames on bundles reserved for mirroring. */
7475 if (in_bundle->mirror_out) {
7476 if (ctx->packet != NULL) {
7477 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7478 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7479 "%s, which is reserved exclusively for mirroring",
7480 ctx->ofproto->up.name, in_bundle->name);
7486 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
7487 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
7490 vlan = input_vid_to_vlan(in_bundle, vid);
7492 /* Look at the output ports to check for destination selections. */
7494 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
7495 ctx->odp_actions->size) {
7496 enum ovs_action_attr type = nl_attr_type(a);
7497 struct ofport_dpif *ofport;
7499 if (type != OVS_ACTION_ATTR_OUTPUT) {
7503 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
7504 if (ofport && ofport->bundle) {
7505 mirrors |= ofport->bundle->dst_mirrors;
7513 /* Restore the original packet before adding the mirror actions. */
7514 ctx->flow = *orig_flow;
7519 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7521 if (!vlan_is_mirrored(m, vlan)) {
7522 mirrors = zero_rightmost_1bit(mirrors);
7526 mirrors &= ~m->dup_mirrors;
7527 ctx->mirrors |= m->dup_mirrors;
7529 output_normal(ctx, m->out, vlan);
7530 } else if (vlan != m->out_vlan
7531 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
7532 struct ofbundle *bundle;
7534 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
7535 if (ofbundle_includes_vlan(bundle, m->out_vlan)
7536 && !bundle->mirror_out) {
7537 output_normal(ctx, bundle, m->out_vlan);
7545 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
7546 uint64_t packets, uint64_t bytes)
7552 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
7555 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7558 /* In normal circumstances 'm' will not be NULL. However,
7559 * if mirrors are reconfigured, we can temporarily get out
7560 * of sync in facet_revalidate(). We could "correct" the
7561 * mirror list before reaching here, but doing that would
7562 * not properly account the traffic stats we've currently
7563 * accumulated for previous mirror configuration. */
7567 m->packet_count += packets;
7568 m->byte_count += bytes;
7572 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
7573 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
7574 * indicate this; newer upstream kernels use gratuitous ARP requests. */
7576 is_gratuitous_arp(const struct flow *flow)
7578 return (flow->dl_type == htons(ETH_TYPE_ARP)
7579 && eth_addr_is_broadcast(flow->dl_dst)
7580 && (flow->nw_proto == ARP_OP_REPLY
7581 || (flow->nw_proto == ARP_OP_REQUEST
7582 && flow->nw_src == flow->nw_dst)));
7586 update_learning_table(struct ofproto_dpif *ofproto,
7587 const struct flow *flow, int vlan,
7588 struct ofbundle *in_bundle)
7590 struct mac_entry *mac;
7592 /* Don't learn the OFPP_NONE port. */
7593 if (in_bundle == &ofpp_none_bundle) {
7597 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
7601 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
7602 if (is_gratuitous_arp(flow)) {
7603 /* We don't want to learn from gratuitous ARP packets that are
7604 * reflected back over bond slaves so we lock the learning table. */
7605 if (!in_bundle->bond) {
7606 mac_entry_set_grat_arp_lock(mac);
7607 } else if (mac_entry_is_grat_arp_locked(mac)) {
7612 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
7613 /* The log messages here could actually be useful in debugging,
7614 * so keep the rate limit relatively high. */
7615 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
7616 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
7617 "on port %s in VLAN %d",
7618 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
7619 in_bundle->name, vlan);
7621 mac->port.p = in_bundle;
7622 tag_set_add(&ofproto->backer->revalidate_set,
7623 mac_learning_changed(ofproto->ml, mac));
7627 static struct ofbundle *
7628 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
7629 bool warn, struct ofport_dpif **in_ofportp)
7631 struct ofport_dpif *ofport;
7633 /* Find the port and bundle for the received packet. */
7634 ofport = get_ofp_port(ofproto, in_port);
7636 *in_ofportp = ofport;
7638 if (ofport && ofport->bundle) {
7639 return ofport->bundle;
7642 /* Special-case OFPP_NONE, which a controller may use as the ingress
7643 * port for traffic that it is sourcing. */
7644 if (in_port == OFPP_NONE) {
7645 return &ofpp_none_bundle;
7648 /* Odd. A few possible reasons here:
7650 * - We deleted a port but there are still a few packets queued up
7653 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
7654 * we don't know about.
7656 * - The ofproto client didn't configure the port as part of a bundle.
7657 * This is particularly likely to happen if a packet was received on the
7658 * port after it was created, but before the client had a chance to
7659 * configure its bundle.
7662 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7664 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
7665 "port %"PRIu16, ofproto->up.name, in_port);
7670 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
7671 * dropped. Returns true if they may be forwarded, false if they should be
7674 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
7675 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
7677 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
7678 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
7679 * checked by input_vid_is_valid().
7681 * May also add tags to '*tags', although the current implementation only does
7682 * so in one special case.
7685 is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port,
7688 struct ofproto_dpif *ofproto = ctx->ofproto;
7689 struct flow *flow = &ctx->flow;
7690 struct ofbundle *in_bundle = in_port->bundle;
7692 /* Drop frames for reserved multicast addresses
7693 * only if forward_bpdu option is absent. */
7694 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
7695 xlate_report(ctx, "packet has reserved destination MAC, dropping");
7699 if (in_bundle->bond) {
7700 struct mac_entry *mac;
7702 switch (bond_check_admissibility(in_bundle->bond, in_port,
7703 flow->dl_dst, &ctx->tags)) {
7708 xlate_report(ctx, "bonding refused admissibility, dropping");
7711 case BV_DROP_IF_MOVED:
7712 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
7713 if (mac && mac->port.p != in_bundle &&
7714 (!is_gratuitous_arp(flow)
7715 || mac_entry_is_grat_arp_locked(mac))) {
7716 xlate_report(ctx, "SLB bond thinks this packet looped back, "
7728 xlate_normal(struct action_xlate_ctx *ctx)
7730 struct ofport_dpif *in_port;
7731 struct ofbundle *in_bundle;
7732 struct mac_entry *mac;
7736 ctx->has_normal = true;
7738 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
7739 ctx->packet != NULL, &in_port);
7741 xlate_report(ctx, "no input bundle, dropping");
7745 /* Drop malformed frames. */
7746 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
7747 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
7748 if (ctx->packet != NULL) {
7749 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7750 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
7751 "VLAN tag received on port %s",
7752 ctx->ofproto->up.name, in_bundle->name);
7754 xlate_report(ctx, "partial VLAN tag, dropping");
7758 /* Drop frames on bundles reserved for mirroring. */
7759 if (in_bundle->mirror_out) {
7760 if (ctx->packet != NULL) {
7761 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7762 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7763 "%s, which is reserved exclusively for mirroring",
7764 ctx->ofproto->up.name, in_bundle->name);
7766 xlate_report(ctx, "input port is mirror output port, dropping");
7771 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
7772 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
7773 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
7776 vlan = input_vid_to_vlan(in_bundle, vid);
7778 /* Check other admissibility requirements. */
7779 if (in_port && !is_admissible(ctx, in_port, vlan)) {
7783 /* Learn source MAC. */
7784 if (ctx->may_learn) {
7785 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
7788 /* Determine output bundle. */
7789 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
7792 if (mac->port.p != in_bundle) {
7793 xlate_report(ctx, "forwarding to learned port");
7794 output_normal(ctx, mac->port.p, vlan);
7796 xlate_report(ctx, "learned port is input port, dropping");
7799 struct ofbundle *bundle;
7801 xlate_report(ctx, "no learned MAC for destination, flooding");
7802 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
7803 if (bundle != in_bundle
7804 && ofbundle_includes_vlan(bundle, vlan)
7805 && bundle->floodable
7806 && !bundle->mirror_out) {
7807 output_normal(ctx, bundle, vlan);
7810 ctx->nf_output_iface = NF_OUT_FLOOD;
7814 /* Optimized flow revalidation.
7816 * It's a difficult problem, in general, to tell which facets need to have
7817 * their actions recalculated whenever the OpenFlow flow table changes. We
7818 * don't try to solve that general problem: for most kinds of OpenFlow flow
7819 * table changes, we recalculate the actions for every facet. This is
7820 * relatively expensive, but it's good enough if the OpenFlow flow table
7821 * doesn't change very often.
7823 * However, we can expect one particular kind of OpenFlow flow table change to
7824 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
7825 * of CPU on revalidating every facet whenever MAC learning modifies the flow
7826 * table, we add a special case that applies to flow tables in which every rule
7827 * has the same form (that is, the same wildcards), except that the table is
7828 * also allowed to have a single "catch-all" flow that matches all packets. We
7829 * optimize this case by tagging all of the facets that resubmit into the table
7830 * and invalidating the same tag whenever a flow changes in that table. The
7831 * end result is that we revalidate just the facets that need it (and sometimes
7832 * a few more, but not all of the facets or even all of the facets that
7833 * resubmit to the table modified by MAC learning). */
7835 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
7836 * into an OpenFlow table with the given 'basis'. */
7838 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
7841 if (minimask_is_catchall(mask)) {
7844 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
7845 return tag_create_deterministic(hash);
7849 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
7850 * taggability of that table.
7852 * This function must be called after *each* change to a flow table. If you
7853 * skip calling it on some changes then the pointer comparisons at the end can
7854 * be invalid if you get unlucky. For example, if a flow removal causes a
7855 * cls_table to be destroyed and then a flow insertion causes a cls_table with
7856 * different wildcards to be created with the same address, then this function
7857 * will incorrectly skip revalidation. */
7859 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
7861 struct table_dpif *table = &ofproto->tables[table_id];
7862 const struct oftable *oftable = &ofproto->up.tables[table_id];
7863 struct cls_table *catchall, *other;
7864 struct cls_table *t;
7866 catchall = other = NULL;
7868 switch (hmap_count(&oftable->cls.tables)) {
7870 /* We could tag this OpenFlow table but it would make the logic a
7871 * little harder and it's a corner case that doesn't seem worth it
7877 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
7878 if (cls_table_is_catchall(t)) {
7880 } else if (!other) {
7883 /* Indicate that we can't tag this by setting both tables to
7884 * NULL. (We know that 'catchall' is already NULL.) */
7891 /* Can't tag this table. */
7895 if (table->catchall_table != catchall || table->other_table != other) {
7896 table->catchall_table = catchall;
7897 table->other_table = other;
7898 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7902 /* Given 'rule' that has changed in some way (either it is a rule being
7903 * inserted, a rule being deleted, or a rule whose actions are being
7904 * modified), marks facets for revalidation to ensure that packets will be
7905 * forwarded correctly according to the new state of the flow table.
7907 * This function must be called after *each* change to a flow table. See
7908 * the comment on table_update_taggable() for more information. */
7910 rule_invalidate(const struct rule_dpif *rule)
7912 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
7914 table_update_taggable(ofproto, rule->up.table_id);
7916 if (!ofproto->backer->need_revalidate) {
7917 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
7919 if (table->other_table && rule->tag) {
7920 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
7922 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7928 set_frag_handling(struct ofproto *ofproto_,
7929 enum ofp_config_flags frag_handling)
7931 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7932 if (frag_handling != OFPC_FRAG_REASM) {
7933 ofproto->backer->need_revalidate = REV_RECONFIGURE;
7941 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
7942 const struct flow *flow,
7943 const struct ofpact *ofpacts, size_t ofpacts_len)
7945 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7946 struct initial_vals initial_vals;
7947 struct odputil_keybuf keybuf;
7948 struct dpif_flow_stats stats;
7952 struct action_xlate_ctx ctx;
7953 uint64_t odp_actions_stub[1024 / 8];
7954 struct ofpbuf odp_actions;
7956 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
7957 odp_flow_key_from_flow(&key, flow,
7958 ofp_port_to_odp_port(ofproto, flow->in_port));
7960 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
7962 initial_vals.vlan_tci = flow->vlan_tci;
7963 initial_vals.tunnel_ip_tos = 0;
7964 action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals, NULL,
7965 packet_get_tcp_flags(packet, flow), packet);
7966 ctx.resubmit_stats = &stats;
7968 ofpbuf_use_stub(&odp_actions,
7969 odp_actions_stub, sizeof odp_actions_stub);
7970 xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions);
7971 dpif_execute(ofproto->backer->dpif, key.data, key.size,
7972 odp_actions.data, odp_actions.size, packet);
7973 ofpbuf_uninit(&odp_actions);
7981 set_netflow(struct ofproto *ofproto_,
7982 const struct netflow_options *netflow_options)
7984 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7986 if (netflow_options) {
7987 if (!ofproto->netflow) {
7988 ofproto->netflow = netflow_create();
7990 return netflow_set_options(ofproto->netflow, netflow_options);
7992 netflow_destroy(ofproto->netflow);
7993 ofproto->netflow = NULL;
7999 get_netflow_ids(const struct ofproto *ofproto_,
8000 uint8_t *engine_type, uint8_t *engine_id)
8002 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
8004 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
8008 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
8010 if (!facet_is_controller_flow(facet) &&
8011 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
8012 struct subfacet *subfacet;
8013 struct ofexpired expired;
8015 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
8016 if (subfacet->path == SF_FAST_PATH) {
8017 struct dpif_flow_stats stats;
8019 subfacet_reinstall(subfacet, &stats);
8020 subfacet_update_stats(subfacet, &stats);
8024 expired.flow = facet->flow;
8025 expired.packet_count = facet->packet_count;
8026 expired.byte_count = facet->byte_count;
8027 expired.used = facet->used;
8028 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
8033 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
8035 struct facet *facet;
8037 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
8038 send_active_timeout(ofproto, facet);
8042 static struct ofproto_dpif *
8043 ofproto_dpif_lookup(const char *name)
8045 struct ofproto_dpif *ofproto;
8047 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
8048 hash_string(name, 0), &all_ofproto_dpifs) {
8049 if (!strcmp(ofproto->up.name, name)) {
8057 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
8058 const char *argv[], void *aux OVS_UNUSED)
8060 struct ofproto_dpif *ofproto;
8063 ofproto = ofproto_dpif_lookup(argv[1]);
8065 unixctl_command_reply_error(conn, "no such bridge");
8068 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
8070 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8071 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
8075 unixctl_command_reply(conn, "table successfully flushed");
8079 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
8080 const char *argv[], void *aux OVS_UNUSED)
8082 struct ds ds = DS_EMPTY_INITIALIZER;
8083 const struct ofproto_dpif *ofproto;
8084 const struct mac_entry *e;
8086 ofproto = ofproto_dpif_lookup(argv[1]);
8088 unixctl_command_reply_error(conn, "no such bridge");
8092 ds_put_cstr(&ds, " port VLAN MAC Age\n");
8093 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
8094 struct ofbundle *bundle = e->port.p;
8095 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
8096 ofbundle_get_a_port(bundle)->odp_port,
8097 e->vlan, ETH_ADDR_ARGS(e->mac),
8098 mac_entry_age(ofproto->ml, e));
8100 unixctl_command_reply(conn, ds_cstr(&ds));
8105 struct action_xlate_ctx ctx;
8111 trace_format_rule(struct ds *result, uint8_t table_id, int level,
8112 const struct rule_dpif *rule)
8114 ds_put_char_multiple(result, '\t', level);
8116 ds_put_cstr(result, "No match\n");
8120 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
8121 table_id, ntohll(rule->up.flow_cookie));
8122 cls_rule_format(&rule->up.cr, result);
8123 ds_put_char(result, '\n');
8125 ds_put_char_multiple(result, '\t', level);
8126 ds_put_cstr(result, "OpenFlow ");
8127 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
8128 ds_put_char(result, '\n');
8132 trace_format_flow(struct ds *result, int level, const char *title,
8133 struct trace_ctx *trace)
8135 ds_put_char_multiple(result, '\t', level);
8136 ds_put_format(result, "%s: ", title);
8137 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
8138 ds_put_cstr(result, "unchanged");
8140 flow_format(result, &trace->ctx.flow);
8141 trace->flow = trace->ctx.flow;
8143 ds_put_char(result, '\n');
8147 trace_format_regs(struct ds *result, int level, const char *title,
8148 struct trace_ctx *trace)
8152 ds_put_char_multiple(result, '\t', level);
8153 ds_put_format(result, "%s:", title);
8154 for (i = 0; i < FLOW_N_REGS; i++) {
8155 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
8157 ds_put_char(result, '\n');
8161 trace_format_odp(struct ds *result, int level, const char *title,
8162 struct trace_ctx *trace)
8164 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
8166 ds_put_char_multiple(result, '\t', level);
8167 ds_put_format(result, "%s: ", title);
8168 format_odp_actions(result, odp_actions->data, odp_actions->size);
8169 ds_put_char(result, '\n');
8173 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
8175 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
8176 struct ds *result = trace->result;
8178 ds_put_char(result, '\n');
8179 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
8180 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
8181 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
8182 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
8186 trace_report(struct action_xlate_ctx *ctx, const char *s)
8188 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
8189 struct ds *result = trace->result;
8191 ds_put_char_multiple(result, '\t', ctx->recurse);
8192 ds_put_cstr(result, s);
8193 ds_put_char(result, '\n');
8197 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
8198 void *aux OVS_UNUSED)
8200 const char *dpname = argv[1];
8201 struct ofproto_dpif *ofproto;
8202 struct ofpbuf odp_key;
8203 struct ofpbuf *packet;
8204 struct initial_vals initial_vals;
8210 ofpbuf_init(&odp_key, 0);
8213 ofproto = ofproto_dpif_lookup(dpname);
8215 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
8219 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
8220 /* ofproto/trace dpname flow [-generate] */
8221 const char *flow_s = argv[2];
8222 const char *generate_s = argv[3];
8224 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
8225 * flow. We guess which type it is based on whether 'flow_s' contains
8226 * an '(', since a datapath flow always contains '(') but an
8227 * OpenFlow-like flow should not (in fact it's allowed but I believe
8228 * that's not documented anywhere).
8230 * An alternative would be to try to parse 'flow_s' both ways, but then
8231 * it would be tricky giving a sensible error message. After all, do
8232 * you just say "syntax error" or do you present both error messages?
8233 * Both choices seem lousy. */
8234 if (strchr(flow_s, '(')) {
8237 /* Convert string to datapath key. */
8238 ofpbuf_init(&odp_key, 0);
8239 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
8241 unixctl_command_reply_error(conn, "Bad flow syntax");
8245 /* The user might have specified the wrong ofproto but within the
8246 * same backer. That's OK, ofproto_receive() can find the right
8248 if (ofproto_receive(ofproto->backer, NULL, odp_key.data,
8249 odp_key.size, &flow, NULL, &ofproto, NULL,
8251 unixctl_command_reply_error(conn, "Invalid flow");
8254 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
8258 error_s = parse_ofp_exact_flow(&flow, argv[2]);
8260 unixctl_command_reply_error(conn, error_s);
8265 initial_vals.vlan_tci = flow.vlan_tci;
8266 initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos;
8269 /* Generate a packet, if requested. */
8271 packet = ofpbuf_new(0);
8272 flow_compose(packet, &flow);
8274 } else if (argc == 7) {
8275 /* ofproto/trace dpname priority tun_id in_port mark packet */
8276 const char *priority_s = argv[2];
8277 const char *tun_id_s = argv[3];
8278 const char *in_port_s = argv[4];
8279 const char *mark_s = argv[5];
8280 const char *packet_s = argv[6];
8281 uint32_t in_port = atoi(in_port_s);
8282 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
8283 uint32_t priority = atoi(priority_s);
8284 uint32_t mark = atoi(mark_s);
8287 msg = eth_from_hex(packet_s, &packet);
8289 unixctl_command_reply_error(conn, msg);
8293 ds_put_cstr(&result, "Packet: ");
8294 s = ofp_packet_to_string(packet->data, packet->size);
8295 ds_put_cstr(&result, s);
8298 flow_extract(packet, priority, mark, NULL, in_port, &flow);
8299 flow.tunnel.tun_id = tun_id;
8300 initial_vals.vlan_tci = flow.vlan_tci;
8301 initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos;
8303 unixctl_command_reply_error(conn, "Bad command syntax");
8307 ofproto_trace(ofproto, &flow, packet, &initial_vals, &result);
8308 unixctl_command_reply(conn, ds_cstr(&result));
8311 ds_destroy(&result);
8312 ofpbuf_delete(packet);
8313 ofpbuf_uninit(&odp_key);
8317 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
8318 const struct ofpbuf *packet,
8319 const struct initial_vals *initial_vals, struct ds *ds)
8321 struct rule_dpif *rule;
8323 ds_put_cstr(ds, "Flow: ");
8324 flow_format(ds, flow);
8325 ds_put_char(ds, '\n');
8327 rule = rule_dpif_lookup(ofproto, flow);
8329 trace_format_rule(ds, 0, 0, rule);
8330 if (rule == ofproto->miss_rule) {
8331 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
8332 } else if (rule == ofproto->no_packet_in_rule) {
8333 ds_put_cstr(ds, "\nNo match, packets dropped because "
8334 "OFPPC_NO_PACKET_IN is set on in_port.\n");
8338 uint64_t odp_actions_stub[1024 / 8];
8339 struct ofpbuf odp_actions;
8341 struct trace_ctx trace;
8344 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
8347 ofpbuf_use_stub(&odp_actions,
8348 odp_actions_stub, sizeof odp_actions_stub);
8349 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_vals,
8350 rule, tcp_flags, packet);
8351 trace.ctx.resubmit_hook = trace_resubmit;
8352 trace.ctx.report_hook = trace_report;
8353 xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len,
8356 ds_put_char(ds, '\n');
8357 trace_format_flow(ds, 0, "Final flow", &trace);
8358 ds_put_cstr(ds, "Datapath actions: ");
8359 format_odp_actions(ds, odp_actions.data, odp_actions.size);
8360 ofpbuf_uninit(&odp_actions);
8362 if (trace.ctx.slow) {
8363 enum slow_path_reason slow;
8365 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
8366 "slow path because it:");
8367 for (slow = trace.ctx.slow; slow; ) {
8368 enum slow_path_reason bit = rightmost_1bit(slow);
8372 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
8375 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
8378 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
8381 ds_put_cstr(ds, "\n\t- Consists of BFD packets.");
8384 ds_put_cstr(ds, "\n\t- Needs in-band special case "
8387 ds_put_cstr(ds, "\n\t (The datapath actions are "
8388 "incomplete--for complete actions, "
8389 "please supply a packet.)");
8392 case SLOW_CONTROLLER:
8393 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
8394 "to the OpenFlow controller.");
8397 ds_put_cstr(ds, "\n\t- Needs more specific matching "
8398 "than the datapath supports.");
8405 if (slow & ~SLOW_MATCH) {
8406 ds_put_cstr(ds, "\nThe datapath actions above do not reflect "
8407 "the special slow-path processing.");
8414 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8415 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8418 unixctl_command_reply(conn, NULL);
8422 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
8423 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
8426 unixctl_command_reply(conn, NULL);
8429 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
8430 * 'reply' describing the results. */
8432 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
8434 struct facet *facet;
8438 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
8439 if (!facet_check_consistency(facet)) {
8444 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
8448 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
8449 ofproto->up.name, errors);
8451 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
8456 ofproto_dpif_self_check(struct unixctl_conn *conn,
8457 int argc, const char *argv[], void *aux OVS_UNUSED)
8459 struct ds reply = DS_EMPTY_INITIALIZER;
8460 struct ofproto_dpif *ofproto;
8463 ofproto = ofproto_dpif_lookup(argv[1]);
8465 unixctl_command_reply_error(conn, "Unknown ofproto (use "
8466 "ofproto/list for help)");
8469 ofproto_dpif_self_check__(ofproto, &reply);
8471 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8472 ofproto_dpif_self_check__(ofproto, &reply);
8476 unixctl_command_reply(conn, ds_cstr(&reply));
8480 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
8481 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
8482 * to destroy 'ofproto_shash' and free the returned value. */
8483 static const struct shash_node **
8484 get_ofprotos(struct shash *ofproto_shash)
8486 const struct ofproto_dpif *ofproto;
8488 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8489 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
8490 shash_add_nocopy(ofproto_shash, name, ofproto);
8493 return shash_sort(ofproto_shash);
8497 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
8498 const char *argv[] OVS_UNUSED,
8499 void *aux OVS_UNUSED)
8501 struct ds ds = DS_EMPTY_INITIALIZER;
8502 struct shash ofproto_shash;
8503 const struct shash_node **sorted_ofprotos;
8506 shash_init(&ofproto_shash);
8507 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8508 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8509 const struct shash_node *node = sorted_ofprotos[i];
8510 ds_put_format(&ds, "%s\n", node->name);
8513 shash_destroy(&ofproto_shash);
8514 free(sorted_ofprotos);
8516 unixctl_command_reply(conn, ds_cstr(&ds));
8521 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
8523 const struct shash_node **ports;
8525 struct avg_subfacet_rates lifetime;
8526 unsigned long long int minutes;
8527 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
8529 minutes = (time_msec() - ofproto->created) / min_ms;
8532 lifetime.add_rate = (double)ofproto->total_subfacet_add_count
8534 lifetime.del_rate = (double)ofproto->total_subfacet_del_count
8537 lifetime.add_rate = 0.0;
8538 lifetime.del_rate = 0.0;
8541 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
8542 dpif_name(ofproto->backer->dpif));
8544 "\tlookups: hit:%"PRIu64" missed:%"PRIu64"\n",
8545 ofproto->n_hit, ofproto->n_missed);
8546 ds_put_format(ds, "\tflows: cur: %zu, avg: %5.3f, max: %d,"
8547 " life span: %llu(ms)\n",
8548 hmap_count(&ofproto->subfacets),
8549 avg_subfacet_count(ofproto),
8550 ofproto->max_n_subfacet,
8551 avg_subfacet_life_span(ofproto));
8552 if (minutes >= 60) {
8553 show_dp_rates(ds, "\t\thourly avg:", &ofproto->hourly);
8555 if (minutes >= 60 * 24) {
8556 show_dp_rates(ds, "\t\tdaily avg:", &ofproto->daily);
8558 show_dp_rates(ds, "\t\toverall avg:", &lifetime);
8560 ports = shash_sort(&ofproto->up.port_by_name);
8561 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
8562 const struct shash_node *node = ports[i];
8563 struct ofport *ofport = node->data;
8564 const char *name = netdev_get_name(ofport->netdev);
8565 const char *type = netdev_get_type(ofport->netdev);
8568 ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port);
8570 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
8571 if (odp_port != OVSP_NONE) {
8572 ds_put_format(ds, "%"PRIu32":", odp_port);
8574 ds_put_cstr(ds, "none:");
8577 if (strcmp(type, "system")) {
8578 struct netdev *netdev;
8581 ds_put_format(ds, " (%s", type);
8583 error = netdev_open(name, type, &netdev);
8588 error = netdev_get_config(netdev, &config);
8590 const struct smap_node **nodes;
8593 nodes = smap_sort(&config);
8594 for (i = 0; i < smap_count(&config); i++) {
8595 const struct smap_node *node = nodes[i];
8596 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
8597 node->key, node->value);
8601 smap_destroy(&config);
8603 netdev_close(netdev);
8605 ds_put_char(ds, ')');
8607 ds_put_char(ds, '\n');
8613 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
8614 const char *argv[], void *aux OVS_UNUSED)
8616 struct ds ds = DS_EMPTY_INITIALIZER;
8617 const struct ofproto_dpif *ofproto;
8621 for (i = 1; i < argc; i++) {
8622 ofproto = ofproto_dpif_lookup(argv[i]);
8624 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
8625 "for help)", argv[i]);
8626 unixctl_command_reply_error(conn, ds_cstr(&ds));
8629 show_dp_format(ofproto, &ds);
8632 struct shash ofproto_shash;
8633 const struct shash_node **sorted_ofprotos;
8636 shash_init(&ofproto_shash);
8637 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8638 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8639 const struct shash_node *node = sorted_ofprotos[i];
8640 show_dp_format(node->data, &ds);
8643 shash_destroy(&ofproto_shash);
8644 free(sorted_ofprotos);
8647 unixctl_command_reply(conn, ds_cstr(&ds));
8652 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
8653 int argc OVS_UNUSED, const char *argv[],
8654 void *aux OVS_UNUSED)
8656 struct ds ds = DS_EMPTY_INITIALIZER;
8657 const struct ofproto_dpif *ofproto;
8658 struct subfacet *subfacet;
8660 ofproto = ofproto_dpif_lookup(argv[1]);
8662 unixctl_command_reply_error(conn, "no such bridge");
8666 update_stats(ofproto->backer);
8668 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
8669 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
8671 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
8672 subfacet->dp_packet_count, subfacet->dp_byte_count);
8673 if (subfacet->used) {
8674 ds_put_format(&ds, "%.3fs",
8675 (time_msec() - subfacet->used) / 1000.0);
8677 ds_put_format(&ds, "never");
8679 if (subfacet->facet->tcp_flags) {
8680 ds_put_cstr(&ds, ", flags:");
8681 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
8684 ds_put_cstr(&ds, ", actions:");
8685 if (subfacet->slow) {
8686 uint64_t slow_path_stub[128 / 8];
8687 const struct nlattr *actions;
8690 compose_slow_path(ofproto, &subfacet->facet->flow, subfacet->slow,
8691 slow_path_stub, sizeof slow_path_stub,
8692 &actions, &actions_len);
8693 format_odp_actions(&ds, actions, actions_len);
8695 format_odp_actions(&ds, subfacet->actions, subfacet->actions_len);
8697 ds_put_char(&ds, '\n');
8700 unixctl_command_reply(conn, ds_cstr(&ds));
8705 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
8706 int argc OVS_UNUSED, const char *argv[],
8707 void *aux OVS_UNUSED)
8709 struct ds ds = DS_EMPTY_INITIALIZER;
8710 struct ofproto_dpif *ofproto;
8712 ofproto = ofproto_dpif_lookup(argv[1]);
8714 unixctl_command_reply_error(conn, "no such bridge");
8718 flush(&ofproto->up);
8720 unixctl_command_reply(conn, ds_cstr(&ds));
8725 ofproto_dpif_unixctl_init(void)
8727 static bool registered;
8733 unixctl_command_register(
8735 "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}",
8736 2, 6, ofproto_unixctl_trace, NULL);
8737 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
8738 ofproto_unixctl_fdb_flush, NULL);
8739 unixctl_command_register("fdb/show", "bridge", 1, 1,
8740 ofproto_unixctl_fdb_show, NULL);
8741 unixctl_command_register("ofproto/clog", "", 0, 0,
8742 ofproto_dpif_clog, NULL);
8743 unixctl_command_register("ofproto/unclog", "", 0, 0,
8744 ofproto_dpif_unclog, NULL);
8745 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
8746 ofproto_dpif_self_check, NULL);
8747 unixctl_command_register("dpif/dump-dps", "", 0, 0,
8748 ofproto_unixctl_dpif_dump_dps, NULL);
8749 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
8750 ofproto_unixctl_dpif_show, NULL);
8751 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
8752 ofproto_unixctl_dpif_dump_flows, NULL);
8753 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
8754 ofproto_unixctl_dpif_del_flows, NULL);
8757 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
8759 * This is deprecated. It is only for compatibility with broken device drivers
8760 * in old versions of Linux that do not properly support VLANs when VLAN
8761 * devices are not used. When broken device drivers are no longer in
8762 * widespread use, we will delete these interfaces. */
8765 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
8767 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
8768 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
8770 if (realdev_ofp_port == ofport->realdev_ofp_port
8771 && vid == ofport->vlandev_vid) {
8775 ofproto->backer->need_revalidate = REV_RECONFIGURE;
8777 if (ofport->realdev_ofp_port) {
8780 if (realdev_ofp_port && ofport->bundle) {
8781 /* vlandevs are enslaved to their realdevs, so they are not allowed to
8782 * themselves be part of a bundle. */
8783 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
8786 ofport->realdev_ofp_port = realdev_ofp_port;
8787 ofport->vlandev_vid = vid;
8789 if (realdev_ofp_port) {
8790 vsp_add(ofport, realdev_ofp_port, vid);
8797 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
8799 return hash_2words(realdev_ofp_port, vid);
8802 /* Returns the ODP port number of the Linux VLAN device that corresponds to
8803 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
8804 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
8805 * it would return the port number of eth0.9.
8807 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
8808 * function just returns its 'realdev_odp_port' argument. */
8810 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
8811 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
8813 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
8814 uint16_t realdev_ofp_port;
8815 int vid = vlan_tci_to_vid(vlan_tci);
8816 const struct vlan_splinter *vsp;
8818 realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port);
8819 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
8820 hash_realdev_vid(realdev_ofp_port, vid),
8821 &ofproto->realdev_vid_map) {
8822 if (vsp->realdev_ofp_port == realdev_ofp_port
8823 && vsp->vid == vid) {
8824 return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port);
8828 return realdev_odp_port;
8831 static struct vlan_splinter *
8832 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
8834 struct vlan_splinter *vsp;
8836 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
8837 &ofproto->vlandev_map) {
8838 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
8846 /* Returns the OpenFlow port number of the "real" device underlying the Linux
8847 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
8848 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
8849 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
8850 * eth0 and store 9 in '*vid'.
8852 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
8853 * VLAN device. Unless VLAN splinters are enabled, this is what this function
8856 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
8857 uint16_t vlandev_ofp_port, int *vid)
8859 if (!hmap_is_empty(&ofproto->vlandev_map)) {
8860 const struct vlan_splinter *vsp;
8862 vsp = vlandev_find(ofproto, vlandev_ofp_port);
8867 return vsp->realdev_ofp_port;
8873 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
8874 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
8875 * 'flow->in_port' to the "real" device backing the VLAN device, sets
8876 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
8877 * always the case unless VLAN splinters are enabled), returns false without
8878 * making any changes. */
8880 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
8885 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
8890 /* Cause the flow to be processed as if it came in on the real device with
8891 * the VLAN device's VLAN ID. */
8892 flow->in_port = realdev;
8893 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
8898 vsp_remove(struct ofport_dpif *port)
8900 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8901 struct vlan_splinter *vsp;
8903 vsp = vlandev_find(ofproto, port->up.ofp_port);
8905 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
8906 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
8909 port->realdev_ofp_port = 0;
8911 VLOG_ERR("missing vlan device record");
8916 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
8918 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8920 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
8921 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
8922 == realdev_ofp_port)) {
8923 struct vlan_splinter *vsp;
8925 vsp = xmalloc(sizeof *vsp);
8926 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
8927 hash_int(port->up.ofp_port, 0));
8928 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
8929 hash_realdev_vid(realdev_ofp_port, vid));
8930 vsp->realdev_ofp_port = realdev_ofp_port;
8931 vsp->vlandev_ofp_port = port->up.ofp_port;
8934 port->realdev_ofp_port = realdev_ofp_port;
8936 VLOG_ERR("duplicate vlan device record");
8941 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
8943 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
8944 return ofport ? ofport->odp_port : OVSP_NONE;
8947 static struct ofport_dpif *
8948 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
8950 struct ofport_dpif *port;
8952 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
8953 hash_int(odp_port, 0),
8954 &backer->odp_to_ofport_map) {
8955 if (port->odp_port == odp_port) {
8964 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
8966 struct ofport_dpif *port;
8968 port = odp_port_to_ofport(ofproto->backer, odp_port);
8969 if (port && &ofproto->up == port->up.ofproto) {
8970 return port->up.ofp_port;
8975 static unsigned long long int
8976 avg_subfacet_life_span(const struct ofproto_dpif *ofproto)
8978 unsigned long long int dc;
8979 unsigned long long int avg;
8981 dc = ofproto->total_subfacet_del_count + ofproto->subfacet_del_count;
8982 avg = dc ? ofproto->total_subfacet_life_span / dc : 0;
8988 avg_subfacet_count(const struct ofproto_dpif *ofproto)
8992 if (ofproto->n_update_stats) {
8993 avg_c = (double)ofproto->total_subfacet_count
8994 / ofproto->n_update_stats;
9001 show_dp_rates(struct ds *ds, const char *heading,
9002 const struct avg_subfacet_rates *rates)
9004 ds_put_format(ds, "%s add rate: %5.3f/min, del rate: %5.3f/min\n",
9005 heading, rates->add_rate, rates->del_rate);
9009 update_max_subfacet_count(struct ofproto_dpif *ofproto)
9011 ofproto->max_n_subfacet = MAX(ofproto->max_n_subfacet,
9012 hmap_count(&ofproto->subfacets));
9015 /* Compute exponentially weighted moving average, adding 'new' as the newest,
9016 * most heavily weighted element. 'base' designates the rate of decay: after
9017 * 'base' further updates, 'new''s weight in the EWMA decays to about 1/e
9020 exp_mavg(double *avg, int base, double new)
9022 *avg = (*avg * (base - 1) + new) / base;
9026 update_moving_averages(struct ofproto_dpif *ofproto)
9028 const int min_ms = 60 * 1000; /* milliseconds in one minute. */
9030 /* Update hourly averages on the minute boundaries. */
9031 if (time_msec() - ofproto->last_minute >= min_ms) {
9032 exp_mavg(&ofproto->hourly.add_rate, 60, ofproto->subfacet_add_count);
9033 exp_mavg(&ofproto->hourly.del_rate, 60, ofproto->subfacet_del_count);
9035 /* Update daily averages on the hour boundaries. */
9036 if ((ofproto->last_minute - ofproto->created) / min_ms % 60 == 59) {
9037 exp_mavg(&ofproto->daily.add_rate, 24, ofproto->hourly.add_rate);
9038 exp_mavg(&ofproto->daily.del_rate, 24, ofproto->hourly.del_rate);
9041 ofproto->total_subfacet_add_count += ofproto->subfacet_add_count;
9042 ofproto->total_subfacet_del_count += ofproto->subfacet_del_count;
9043 ofproto->subfacet_add_count = 0;
9044 ofproto->subfacet_del_count = 0;
9045 ofproto->last_minute += min_ms;
9050 dpif_stats_update_hit_count(struct ofproto_dpif *ofproto, uint64_t delta)
9052 ofproto->n_hit += delta;
9055 const struct ofproto_class ofproto_dpif_class = {
9090 port_is_lacp_current,
9091 NULL, /* rule_choose_table */
9098 rule_modify_actions,
9112 get_stp_port_status,
9119 is_mirror_output_bundle,
9120 forward_bpdu_changed,
9121 set_mac_table_config,