2 * Copyright (c) 2009, 2010, 2011, 2012, 2013 Nicira, Inc.
4 * Licensed under the Apache License, Version 2.0 (the "License");
5 * you may not use this file except in compliance with the License.
6 * You may obtain a copy of the License at:
8 * http://www.apache.org/licenses/LICENSE-2.0
10 * Unless required by applicable law or agreed to in writing, software
11 * distributed under the License is distributed on an "AS IS" BASIS,
12 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13 * See the License for the specific language governing permissions and
14 * limitations under the License.
19 #include "ofproto/ofproto-provider.h"
26 #include "byte-order.h"
31 #include "dynamic-string.h"
32 #include "fail-open.h"
36 #include "mac-learning.h"
37 #include "meta-flow.h"
38 #include "multipath.h"
39 #include "netdev-vport.h"
46 #include "ofp-actions.h"
47 #include "ofp-parse.h"
48 #include "ofp-print.h"
49 #include "ofproto-dpif-governor.h"
50 #include "ofproto-dpif-sflow.h"
51 #include "poll-loop.h"
56 #include "unaligned.h"
58 #include "vlan-bitmap.h"
61 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
63 COVERAGE_DEFINE(ofproto_dpif_expired);
64 COVERAGE_DEFINE(ofproto_dpif_xlate);
65 COVERAGE_DEFINE(facet_changed_rule);
66 COVERAGE_DEFINE(facet_revalidate);
67 COVERAGE_DEFINE(facet_unexpected);
68 COVERAGE_DEFINE(facet_suppress);
70 /* Maximum depth of flow table recursion (due to resubmit actions) in a
71 * flow translation. */
72 #define MAX_RESUBMIT_RECURSION 64
74 /* Number of implemented OpenFlow tables. */
75 enum { N_TABLES = 255 };
76 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
77 BUILD_ASSERT_DECL(N_TABLES >= 2 && N_TABLES <= 255);
88 * - Do include packets and bytes from facets that have been deleted or
89 * whose own statistics have been folded into the rule.
91 * - Do include packets and bytes sent "by hand" that were accounted to
92 * the rule without any facet being involved (this is a rare corner
93 * case in rule_execute()).
95 * - Do not include packet or bytes that can be obtained from any facet's
96 * packet_count or byte_count member or that can be obtained from the
97 * datapath by, e.g., dpif_flow_get() for any subfacet.
99 uint64_t packet_count; /* Number of packets received. */
100 uint64_t byte_count; /* Number of bytes received. */
102 tag_type tag; /* Caches rule_calculate_tag() result. */
104 struct list facets; /* List of "struct facet"s. */
107 static struct rule_dpif *rule_dpif_cast(const struct rule *rule)
109 return rule ? CONTAINER_OF(rule, struct rule_dpif, up) : NULL;
112 static struct rule_dpif *rule_dpif_lookup(struct ofproto_dpif *,
113 const struct flow *);
114 static struct rule_dpif *rule_dpif_lookup__(struct ofproto_dpif *,
117 static struct rule_dpif *rule_dpif_miss_rule(struct ofproto_dpif *ofproto,
118 const struct flow *flow);
120 static void rule_credit_stats(struct rule_dpif *,
121 const struct dpif_flow_stats *);
122 static void flow_push_stats(struct rule_dpif *, const struct flow *,
123 const struct dpif_flow_stats *);
124 static tag_type rule_calculate_tag(const struct flow *,
125 const struct minimask *, uint32_t basis);
126 static void rule_invalidate(const struct rule_dpif *);
128 #define MAX_MIRRORS 32
129 typedef uint32_t mirror_mask_t;
130 #define MIRROR_MASK_C(X) UINT32_C(X)
131 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
133 struct ofproto_dpif *ofproto; /* Owning ofproto. */
134 size_t idx; /* In ofproto's "mirrors" array. */
135 void *aux; /* Key supplied by ofproto's client. */
136 char *name; /* Identifier for log messages. */
138 /* Selection criteria. */
139 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
140 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
141 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
143 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
144 struct ofbundle *out; /* Output port or NULL. */
145 int out_vlan; /* Output VLAN or -1. */
146 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
149 int64_t packet_count; /* Number of packets sent. */
150 int64_t byte_count; /* Number of bytes sent. */
153 static void mirror_destroy(struct ofmirror *);
154 static void update_mirror_stats(struct ofproto_dpif *ofproto,
155 mirror_mask_t mirrors,
156 uint64_t packets, uint64_t bytes);
159 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
160 struct ofproto_dpif *ofproto; /* Owning ofproto. */
161 void *aux; /* Key supplied by ofproto's client. */
162 char *name; /* Identifier for log messages. */
165 struct list ports; /* Contains "struct ofport"s. */
166 enum port_vlan_mode vlan_mode; /* VLAN mode */
167 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
168 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
169 * NULL if all VLANs are trunked. */
170 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
171 struct bond *bond; /* Nonnull iff more than one port. */
172 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
175 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
177 /* Port mirroring info. */
178 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
179 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
180 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
183 static void bundle_remove(struct ofport *);
184 static void bundle_update(struct ofbundle *);
185 static void bundle_destroy(struct ofbundle *);
186 static void bundle_del_port(struct ofport_dpif *);
187 static void bundle_run(struct ofbundle *);
188 static void bundle_wait(struct ofbundle *);
189 static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *,
190 uint16_t in_port, bool warn,
191 struct ofport_dpif **in_ofportp);
193 /* A controller may use OFPP_NONE as the ingress port to indicate that
194 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
195 * when an input bundle is needed for validation (e.g., mirroring or
196 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
197 * any 'port' structs, so care must be taken when dealing with it. */
198 static struct ofbundle ofpp_none_bundle = {
200 .vlan_mode = PORT_VLAN_TRUNK
203 static void stp_run(struct ofproto_dpif *ofproto);
204 static void stp_wait(struct ofproto_dpif *ofproto);
205 static int set_stp_port(struct ofport *,
206 const struct ofproto_port_stp_settings *);
208 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
210 struct action_xlate_ctx {
211 /* action_xlate_ctx_init() initializes these members. */
214 struct ofproto_dpif *ofproto;
216 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
217 * this flow when actions change header fields. */
220 /* The packet corresponding to 'flow', or a null pointer if we are
221 * revalidating without a packet to refer to. */
222 const struct ofpbuf *packet;
224 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
225 * actions update the flow table?
227 * We want to update these tables if we are actually processing a packet,
228 * or if we are accounting for packets that the datapath has processed, but
229 * not if we are just revalidating. */
232 /* The rule that we are currently translating, or NULL. */
233 struct rule_dpif *rule;
235 /* Union of the set of TCP flags seen so far in this flow. (Used only by
236 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
240 /* If nonnull, flow translation calls this function just before executing a
241 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
242 * when the recursion depth is exceeded.
244 * 'rule' is the rule being submitted into. It will be null if the
245 * resubmit or OFPP_TABLE action didn't find a matching rule.
247 * This is normally null so the client has to set it manually after
248 * calling action_xlate_ctx_init(). */
249 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule);
251 /* If nonnull, flow translation calls this function to report some
252 * significant decision, e.g. to explain why OFPP_NORMAL translation
253 * dropped a packet. */
254 void (*report_hook)(struct action_xlate_ctx *, const char *s);
256 /* If nonnull, flow translation credits the specified statistics to each
257 * rule reached through a resubmit or OFPP_TABLE action.
259 * This is normally null so the client has to set it manually after
260 * calling action_xlate_ctx_init(). */
261 const struct dpif_flow_stats *resubmit_stats;
263 /* xlate_actions() initializes and uses these members. The client might want
264 * to look at them after it returns. */
266 struct ofpbuf *odp_actions; /* Datapath actions. */
267 tag_type tags; /* Tags associated with actions. */
268 enum slow_path_reason slow; /* 0 if fast path may be used. */
269 bool has_learn; /* Actions include NXAST_LEARN? */
270 bool has_normal; /* Actions output to OFPP_NORMAL? */
271 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
272 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
273 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
275 /* xlate_actions() initializes and uses these members, but the client has no
276 * reason to look at them. */
278 int recurse; /* Recursion level, via xlate_table_action. */
279 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
280 struct flow base_flow; /* Flow at the last commit. */
281 uint32_t orig_skb_priority; /* Priority when packet arrived. */
282 uint8_t table_id; /* OpenFlow table ID where flow was found. */
283 uint32_t sflow_n_outputs; /* Number of output ports. */
284 uint32_t sflow_odp_port; /* Output port for composing sFlow action. */
285 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
286 bool exit; /* No further actions should be processed. */
287 struct flow orig_flow; /* Copy of original flow. */
290 static void action_xlate_ctx_init(struct action_xlate_ctx *,
291 struct ofproto_dpif *, const struct flow *,
292 ovs_be16 initial_tci, struct rule_dpif *,
293 uint8_t tcp_flags, const struct ofpbuf *);
294 static void xlate_actions(struct action_xlate_ctx *,
295 const struct ofpact *ofpacts, size_t ofpacts_len,
296 struct ofpbuf *odp_actions);
297 static void xlate_actions_for_side_effects(struct action_xlate_ctx *,
298 const struct ofpact *ofpacts,
300 static void xlate_table_action(struct action_xlate_ctx *, uint16_t in_port,
301 uint8_t table_id, bool may_packet_in);
303 static size_t put_userspace_action(const struct ofproto_dpif *,
304 struct ofpbuf *odp_actions,
306 const union user_action_cookie *);
308 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
309 enum slow_path_reason,
310 uint64_t *stub, size_t stub_size,
311 const struct nlattr **actionsp,
312 size_t *actions_lenp);
314 static void xlate_report(struct action_xlate_ctx *ctx, const char *s);
316 /* A subfacet (see "struct subfacet" below) has three possible installation
319 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
320 * case just after the subfacet is created, just before the subfacet is
321 * destroyed, or if the datapath returns an error when we try to install a
324 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
326 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
327 * ofproto_dpif is installed in the datapath.
330 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
331 SF_FAST_PATH, /* Full actions are installed. */
332 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
335 static const char *subfacet_path_to_string(enum subfacet_path);
337 /* A dpif flow and actions associated with a facet.
339 * See also the large comment on struct facet. */
342 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
343 struct list list_node; /* In struct facet's 'facets' list. */
344 struct facet *facet; /* Owning facet. */
348 * To save memory in the common case, 'key' is NULL if 'key_fitness' is
349 * ODP_FIT_PERFECT, that is, odp_flow_key_from_flow() can accurately
350 * regenerate the ODP flow key from ->facet->flow. */
351 enum odp_key_fitness key_fitness;
355 long long int used; /* Time last used; time created if not used. */
357 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
358 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
362 * These should be essentially identical for every subfacet in a facet, but
363 * may differ in trivial ways due to VLAN splinters. */
364 size_t actions_len; /* Number of bytes in actions[]. */
365 struct nlattr *actions; /* Datapath actions. */
367 enum slow_path_reason slow; /* 0 if fast path may be used. */
368 enum subfacet_path path; /* Installed in datapath? */
370 /* This value is normally the same as ->facet->flow.vlan_tci. Only VLAN
371 * splinters can cause it to differ. This value should be removed when
372 * the VLAN splinters feature is no longer needed. */
373 ovs_be16 initial_tci; /* Initial VLAN TCI value. */
375 /* Datapath port the packet arrived on. This is needed to remove
376 * flows for ports that are no longer part of the bridge. Since the
377 * flow definition only has the OpenFlow port number and the port is
378 * no longer part of the bridge, we can't determine the datapath port
379 * number needed to delete the flow from the datapath. */
380 uint32_t odp_in_port;
383 #define SUBFACET_DESTROY_MAX_BATCH 50
385 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
387 static struct subfacet *subfacet_find(struct ofproto_dpif *,
388 const struct nlattr *key, size_t key_len,
390 const struct flow *flow);
391 static void subfacet_destroy(struct subfacet *);
392 static void subfacet_destroy__(struct subfacet *);
393 static void subfacet_destroy_batch(struct ofproto_dpif *,
394 struct subfacet **, int n);
395 static void subfacet_get_key(struct subfacet *, struct odputil_keybuf *,
397 static void subfacet_reset_dp_stats(struct subfacet *,
398 struct dpif_flow_stats *);
399 static void subfacet_update_time(struct subfacet *, long long int used);
400 static void subfacet_update_stats(struct subfacet *,
401 const struct dpif_flow_stats *);
402 static void subfacet_make_actions(struct subfacet *,
403 const struct ofpbuf *packet,
404 struct ofpbuf *odp_actions);
405 static int subfacet_install(struct subfacet *,
406 const struct nlattr *actions, size_t actions_len,
407 struct dpif_flow_stats *, enum slow_path_reason);
408 static void subfacet_uninstall(struct subfacet *);
410 static enum subfacet_path subfacet_want_path(enum slow_path_reason);
412 /* An exact-match instantiation of an OpenFlow flow.
414 * A facet associates a "struct flow", which represents the Open vSwitch
415 * userspace idea of an exact-match flow, with one or more subfacets. Each
416 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
417 * the facet. When the kernel module (or other dpif implementation) and Open
418 * vSwitch userspace agree on the definition of a flow key, there is exactly
419 * one subfacet per facet. If the dpif implementation supports more-specific
420 * flow matching than userspace, however, a facet can have more than one
421 * subfacet, each of which corresponds to some distinction in flow that
422 * userspace simply doesn't understand.
424 * Flow expiration works in terms of subfacets, so a facet must have at least
425 * one subfacet or it will never expire, leaking memory. */
428 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
429 struct list list_node; /* In owning rule's 'facets' list. */
430 struct rule_dpif *rule; /* Owning rule. */
433 struct list subfacets;
434 long long int used; /* Time last used; time created if not used. */
441 * - Do include packets and bytes sent "by hand", e.g. with
444 * - Do include packets and bytes that were obtained from the datapath
445 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
446 * DPIF_FP_ZERO_STATS).
448 * - Do not include packets or bytes that can be obtained from the
449 * datapath for any existing subfacet.
451 uint64_t packet_count; /* Number of packets received. */
452 uint64_t byte_count; /* Number of bytes received. */
454 /* Resubmit statistics. */
455 uint64_t prev_packet_count; /* Number of packets from last stats push. */
456 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
457 long long int prev_used; /* Used time from last stats push. */
460 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
461 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
462 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
464 /* Properties of datapath actions.
466 * Every subfacet has its own actions because actions can differ slightly
467 * between splintered and non-splintered subfacets due to the VLAN tag
468 * being initially different (present vs. absent). All of them have these
469 * properties in common so we just store one copy of them here. */
470 bool has_learn; /* Actions include NXAST_LEARN? */
471 bool has_normal; /* Actions output to OFPP_NORMAL? */
472 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
473 tag_type tags; /* Tags that would require revalidation. */
474 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
476 /* Storage for a single subfacet, to reduce malloc() time and space
477 * overhead. (A facet always has at least one subfacet and in the common
478 * case has exactly one subfacet.) */
479 struct subfacet one_subfacet;
482 static struct facet *facet_create(struct rule_dpif *,
483 const struct flow *, uint32_t hash);
484 static void facet_remove(struct facet *);
485 static void facet_free(struct facet *);
487 static struct facet *facet_find(struct ofproto_dpif *,
488 const struct flow *, uint32_t hash);
489 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
490 const struct flow *, uint32_t hash);
491 static void facet_revalidate(struct facet *);
492 static bool facet_check_consistency(struct facet *);
494 static void facet_flush_stats(struct facet *);
496 static void facet_update_time(struct facet *, long long int used);
497 static void facet_reset_counters(struct facet *);
498 static void facet_push_stats(struct facet *);
499 static void facet_learn(struct facet *);
500 static void facet_account(struct facet *);
502 static bool facet_is_controller_flow(struct facet *);
505 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
509 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
510 struct list bundle_node; /* In struct ofbundle's "ports" list. */
511 struct cfm *cfm; /* Connectivity Fault Management, if any. */
512 tag_type tag; /* Tag associated with this port. */
513 uint32_t bond_stable_id; /* stable_id to use as bond slave, or 0. */
514 bool may_enable; /* May be enabled in bonds. */
515 long long int carrier_seq; /* Carrier status changes. */
516 struct tnl_port *tnl_port; /* Tunnel handle, or null. */
519 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
520 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
521 long long int stp_state_entered;
523 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
525 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
527 * This is deprecated. It is only for compatibility with broken device
528 * drivers in old versions of Linux that do not properly support VLANs when
529 * VLAN devices are not used. When broken device drivers are no longer in
530 * widespread use, we will delete these interfaces. */
531 uint16_t realdev_ofp_port;
535 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
536 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
537 * traffic egressing the 'ofport' with that priority should be marked with. */
538 struct priority_to_dscp {
539 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
540 uint32_t priority; /* Priority of this queue (see struct flow). */
542 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
545 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
547 * This is deprecated. It is only for compatibility with broken device drivers
548 * in old versions of Linux that do not properly support VLANs when VLAN
549 * devices are not used. When broken device drivers are no longer in
550 * widespread use, we will delete these interfaces. */
551 struct vlan_splinter {
552 struct hmap_node realdev_vid_node;
553 struct hmap_node vlandev_node;
554 uint16_t realdev_ofp_port;
555 uint16_t vlandev_ofp_port;
559 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
560 uint32_t realdev, ovs_be16 vlan_tci);
561 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
562 static void vsp_remove(struct ofport_dpif *);
563 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
565 static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *,
567 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
570 static struct ofport_dpif *
571 ofport_dpif_cast(const struct ofport *ofport)
573 ovs_assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
574 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
577 static void port_run(struct ofport_dpif *);
578 static void port_run_fast(struct ofport_dpif *);
579 static void port_wait(struct ofport_dpif *);
580 static int set_cfm(struct ofport *, const struct cfm_settings *);
581 static void ofport_clear_priorities(struct ofport_dpif *);
583 struct dpif_completion {
584 struct list list_node;
585 struct ofoperation *op;
588 /* Extra information about a classifier table.
589 * Currently used just for optimized flow revalidation. */
591 /* If either of these is nonnull, then this table has a form that allows
592 * flows to be tagged to avoid revalidating most flows for the most common
593 * kinds of flow table changes. */
594 struct cls_table *catchall_table; /* Table that wildcards all fields. */
595 struct cls_table *other_table; /* Table with any other wildcard set. */
596 uint32_t basis; /* Keeps each table's tags separate. */
599 /* Reasons that we might need to revalidate every facet, and corresponding
602 * A value of 0 means that there is no need to revalidate.
604 * It would be nice to have some cleaner way to integrate with coverage
605 * counters, but with only a few reasons I guess this is good enough for
607 enum revalidate_reason {
608 REV_RECONFIGURE = 1, /* Switch configuration changed. */
609 REV_STP, /* Spanning tree protocol port status change. */
610 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
611 REV_FLOW_TABLE, /* Flow table changed. */
612 REV_INCONSISTENCY /* Facet self-check failed. */
614 COVERAGE_DEFINE(rev_reconfigure);
615 COVERAGE_DEFINE(rev_stp);
616 COVERAGE_DEFINE(rev_port_toggled);
617 COVERAGE_DEFINE(rev_flow_table);
618 COVERAGE_DEFINE(rev_inconsistency);
620 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
621 * These are datapath flows which have no associated ofproto, if they did we
622 * would use facets. */
624 struct hmap_node hmap_node;
629 /* All datapaths of a given type share a single dpif backer instance. */
634 struct timer next_expiration;
635 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
637 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
639 /* Facet revalidation flags applying to facets which use this backer. */
640 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
641 struct tag_set revalidate_set; /* Revalidate only matching facets. */
643 struct hmap drop_keys; /* Set of dropped odp keys. */
646 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
647 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
649 static void drop_key_clear(struct dpif_backer *);
650 static struct ofport_dpif *
651 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
653 struct ofproto_dpif {
654 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
656 struct dpif_backer *backer;
658 /* Special OpenFlow rules. */
659 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
660 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
666 struct netflow *netflow;
667 struct dpif_sflow *sflow;
668 struct hmap bundles; /* Contains "struct ofbundle"s. */
669 struct mac_learning *ml;
670 struct ofmirror *mirrors[MAX_MIRRORS];
672 bool has_bonded_bundles;
676 struct hmap subfacets;
677 struct governor *governor;
680 struct table_dpif tables[N_TABLES];
682 /* Support for debugging async flow mods. */
683 struct list completions;
685 bool has_bundle_action; /* True when the first bundle action appears. */
686 struct netdev_stats stats; /* To account packets generated and consumed in
691 long long int stp_last_tick;
693 /* VLAN splinters. */
694 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
695 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
698 struct sset ports; /* Set of standard port names. */
699 struct sset ghost_ports; /* Ports with no datapath port. */
700 struct sset port_poll_set; /* Queued names for port_poll() reply. */
701 int port_poll_errno; /* Last errno for port_poll() reply. */
704 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
705 * for debugging the asynchronous flow_mod implementation.) */
708 /* All existing ofproto_dpif instances, indexed by ->up.name. */
709 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
711 static void ofproto_dpif_unixctl_init(void);
713 static struct ofproto_dpif *
714 ofproto_dpif_cast(const struct ofproto *ofproto)
716 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
717 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
720 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
722 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
724 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
725 const struct ofpbuf *, ovs_be16 initial_tci,
728 /* Packet processing. */
729 static void update_learning_table(struct ofproto_dpif *,
730 const struct flow *, int vlan,
733 #define FLOW_MISS_MAX_BATCH 50
734 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
736 /* Flow expiration. */
737 static int expire(struct dpif_backer *);
740 static void send_netflow_active_timeouts(struct ofproto_dpif *);
743 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
744 static size_t compose_sflow_action(const struct ofproto_dpif *,
745 struct ofpbuf *odp_actions,
746 const struct flow *, uint32_t odp_port);
747 static void add_mirror_actions(struct action_xlate_ctx *ctx,
748 const struct flow *flow);
749 /* Global variables. */
750 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
752 /* Initial mappings of port to bridge mappings. */
753 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
755 /* Factory functions. */
758 init(const struct shash *iface_hints)
760 struct shash_node *node;
762 /* Make a local copy, since we don't own 'iface_hints' elements. */
763 SHASH_FOR_EACH(node, iface_hints) {
764 const struct iface_hint *orig_hint = node->data;
765 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
767 new_hint->br_name = xstrdup(orig_hint->br_name);
768 new_hint->br_type = xstrdup(orig_hint->br_type);
769 new_hint->ofp_port = orig_hint->ofp_port;
771 shash_add(&init_ofp_ports, node->name, new_hint);
776 enumerate_types(struct sset *types)
778 dp_enumerate_types(types);
782 enumerate_names(const char *type, struct sset *names)
784 struct ofproto_dpif *ofproto;
787 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
788 if (strcmp(type, ofproto->up.type)) {
791 sset_add(names, ofproto->up.name);
798 del(const char *type, const char *name)
803 error = dpif_open(name, type, &dpif);
805 error = dpif_delete(dpif);
812 port_open_type(const char *datapath_type, const char *port_type)
814 return dpif_port_open_type(datapath_type, port_type);
817 /* Type functions. */
819 static struct ofproto_dpif *
820 lookup_ofproto_dpif_by_port_name(const char *name)
822 struct ofproto_dpif *ofproto;
824 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
825 if (sset_contains(&ofproto->ports, name)) {
834 type_run(const char *type)
836 struct dpif_backer *backer;
840 backer = shash_find_data(&all_dpif_backers, type);
842 /* This is not necessarily a problem, since backers are only
843 * created on demand. */
847 dpif_run(backer->dpif);
849 if (backer->need_revalidate
850 || !tag_set_is_empty(&backer->revalidate_set)) {
851 struct tag_set revalidate_set = backer->revalidate_set;
852 bool need_revalidate = backer->need_revalidate;
853 struct ofproto_dpif *ofproto;
854 struct simap_node *node;
855 struct simap tmp_backers;
857 /* Handle tunnel garbage collection. */
858 simap_init(&tmp_backers);
859 simap_swap(&backer->tnl_backers, &tmp_backers);
861 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
862 struct ofport_dpif *iter;
864 if (backer != ofproto->backer) {
868 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
871 if (!iter->tnl_port) {
875 dp_port = netdev_vport_get_dpif_port(iter->up.netdev);
876 node = simap_find(&tmp_backers, dp_port);
878 simap_put(&backer->tnl_backers, dp_port, node->data);
879 simap_delete(&tmp_backers, node);
880 node = simap_find(&backer->tnl_backers, dp_port);
882 node = simap_find(&backer->tnl_backers, dp_port);
884 uint32_t odp_port = UINT32_MAX;
886 if (!dpif_port_add(backer->dpif, iter->up.netdev,
888 simap_put(&backer->tnl_backers, dp_port, odp_port);
889 node = simap_find(&backer->tnl_backers, dp_port);
894 iter->odp_port = node ? node->data : OVSP_NONE;
895 if (tnl_port_reconfigure(&iter->up, iter->odp_port,
897 backer->need_revalidate = REV_RECONFIGURE;
902 SIMAP_FOR_EACH (node, &tmp_backers) {
903 dpif_port_del(backer->dpif, node->data);
905 simap_destroy(&tmp_backers);
907 switch (backer->need_revalidate) {
908 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
909 case REV_STP: COVERAGE_INC(rev_stp); break;
910 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
911 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
912 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
915 if (backer->need_revalidate) {
916 /* Clear the drop_keys in case we should now be accepting some
917 * formerly dropped flows. */
918 drop_key_clear(backer);
921 /* Clear the revalidation flags. */
922 tag_set_init(&backer->revalidate_set);
923 backer->need_revalidate = 0;
925 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
928 if (ofproto->backer != backer) {
932 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
934 || tag_set_intersects(&revalidate_set, facet->tags)) {
935 facet_revalidate(facet);
941 if (timer_expired(&backer->next_expiration)) {
942 int delay = expire(backer);
943 timer_set_duration(&backer->next_expiration, delay);
946 /* Check for port changes in the dpif. */
947 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
948 struct ofproto_dpif *ofproto;
949 struct dpif_port port;
951 /* Don't report on the datapath's device. */
952 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
956 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
957 &all_ofproto_dpifs) {
958 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
963 ofproto = lookup_ofproto_dpif_by_port_name(devname);
964 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
965 /* The port was removed. If we know the datapath,
966 * report it through poll_set(). If we don't, it may be
967 * notifying us of a removal we initiated, so ignore it.
968 * If there's a pending ENOBUFS, let it stand, since
969 * everything will be reevaluated. */
970 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
971 sset_add(&ofproto->port_poll_set, devname);
972 ofproto->port_poll_errno = 0;
974 } else if (!ofproto) {
975 /* The port was added, but we don't know with which
976 * ofproto we should associate it. Delete it. */
977 dpif_port_del(backer->dpif, port.port_no);
979 dpif_port_destroy(&port);
985 if (error != EAGAIN) {
986 struct ofproto_dpif *ofproto;
988 /* There was some sort of error, so propagate it to all
989 * ofprotos that use this backer. */
990 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
991 &all_ofproto_dpifs) {
992 if (ofproto->backer == backer) {
993 sset_clear(&ofproto->port_poll_set);
994 ofproto->port_poll_errno = error;
1003 type_run_fast(const char *type)
1005 struct dpif_backer *backer;
1008 backer = shash_find_data(&all_dpif_backers, type);
1010 /* This is not necessarily a problem, since backers are only
1011 * created on demand. */
1015 /* Handle one or more batches of upcalls, until there's nothing left to do
1016 * or until we do a fixed total amount of work.
1018 * We do work in batches because it can be much cheaper to set up a number
1019 * of flows and fire off their patches all at once. We do multiple batches
1020 * because in some cases handling a packet can cause another packet to be
1021 * queued almost immediately as part of the return flow. Both
1022 * optimizations can make major improvements on some benchmarks and
1023 * presumably for real traffic as well. */
1025 while (work < FLOW_MISS_MAX_BATCH) {
1026 int retval = handle_upcalls(backer, FLOW_MISS_MAX_BATCH - work);
1037 type_wait(const char *type)
1039 struct dpif_backer *backer;
1041 backer = shash_find_data(&all_dpif_backers, type);
1043 /* This is not necessarily a problem, since backers are only
1044 * created on demand. */
1048 timer_wait(&backer->next_expiration);
1051 /* Basic life-cycle. */
1053 static int add_internal_flows(struct ofproto_dpif *);
1055 static struct ofproto *
1058 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1059 return &ofproto->up;
1063 dealloc(struct ofproto *ofproto_)
1065 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1070 close_dpif_backer(struct dpif_backer *backer)
1072 struct shash_node *node;
1074 ovs_assert(backer->refcount > 0);
1076 if (--backer->refcount) {
1080 drop_key_clear(backer);
1081 hmap_destroy(&backer->drop_keys);
1083 simap_destroy(&backer->tnl_backers);
1084 hmap_destroy(&backer->odp_to_ofport_map);
1085 node = shash_find(&all_dpif_backers, backer->type);
1087 shash_delete(&all_dpif_backers, node);
1088 dpif_close(backer->dpif);
1093 /* Datapath port slated for removal from datapath. */
1094 struct odp_garbage {
1095 struct list list_node;
1100 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1102 struct dpif_backer *backer;
1103 struct dpif_port_dump port_dump;
1104 struct dpif_port port;
1105 struct shash_node *node;
1106 struct list garbage_list;
1107 struct odp_garbage *garbage, *next;
1113 backer = shash_find_data(&all_dpif_backers, type);
1120 backer_name = xasprintf("ovs-%s", type);
1122 /* Remove any existing datapaths, since we assume we're the only
1123 * userspace controlling the datapath. */
1125 dp_enumerate_names(type, &names);
1126 SSET_FOR_EACH(name, &names) {
1127 struct dpif *old_dpif;
1129 /* Don't remove our backer if it exists. */
1130 if (!strcmp(name, backer_name)) {
1134 if (dpif_open(name, type, &old_dpif)) {
1135 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1137 dpif_delete(old_dpif);
1138 dpif_close(old_dpif);
1141 sset_destroy(&names);
1143 backer = xmalloc(sizeof *backer);
1145 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1148 VLOG_ERR("failed to open datapath of type %s: %s", type,
1154 backer->type = xstrdup(type);
1155 backer->refcount = 1;
1156 hmap_init(&backer->odp_to_ofport_map);
1157 hmap_init(&backer->drop_keys);
1158 timer_set_duration(&backer->next_expiration, 1000);
1159 backer->need_revalidate = 0;
1160 simap_init(&backer->tnl_backers);
1161 tag_set_init(&backer->revalidate_set);
1164 dpif_flow_flush(backer->dpif);
1166 /* Loop through the ports already on the datapath and remove any
1167 * that we don't need anymore. */
1168 list_init(&garbage_list);
1169 dpif_port_dump_start(&port_dump, backer->dpif);
1170 while (dpif_port_dump_next(&port_dump, &port)) {
1171 node = shash_find(&init_ofp_ports, port.name);
1172 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1173 garbage = xmalloc(sizeof *garbage);
1174 garbage->odp_port = port.port_no;
1175 list_push_front(&garbage_list, &garbage->list_node);
1178 dpif_port_dump_done(&port_dump);
1180 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1181 dpif_port_del(backer->dpif, garbage->odp_port);
1182 list_remove(&garbage->list_node);
1186 shash_add(&all_dpif_backers, type, backer);
1188 error = dpif_recv_set(backer->dpif, true);
1190 VLOG_ERR("failed to listen on datapath of type %s: %s",
1191 type, strerror(error));
1192 close_dpif_backer(backer);
1200 construct(struct ofproto *ofproto_)
1202 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1203 struct shash_node *node, *next;
1208 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1213 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1214 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1216 ofproto->n_matches = 0;
1218 ofproto->netflow = NULL;
1219 ofproto->sflow = NULL;
1220 ofproto->stp = NULL;
1221 hmap_init(&ofproto->bundles);
1222 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1223 for (i = 0; i < MAX_MIRRORS; i++) {
1224 ofproto->mirrors[i] = NULL;
1226 ofproto->has_bonded_bundles = false;
1228 hmap_init(&ofproto->facets);
1229 hmap_init(&ofproto->subfacets);
1230 ofproto->governor = NULL;
1232 for (i = 0; i < N_TABLES; i++) {
1233 struct table_dpif *table = &ofproto->tables[i];
1235 table->catchall_table = NULL;
1236 table->other_table = NULL;
1237 table->basis = random_uint32();
1240 list_init(&ofproto->completions);
1242 ofproto_dpif_unixctl_init();
1244 ofproto->has_mirrors = false;
1245 ofproto->has_bundle_action = false;
1247 hmap_init(&ofproto->vlandev_map);
1248 hmap_init(&ofproto->realdev_vid_map);
1250 sset_init(&ofproto->ports);
1251 sset_init(&ofproto->ghost_ports);
1252 sset_init(&ofproto->port_poll_set);
1253 ofproto->port_poll_errno = 0;
1255 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1256 struct iface_hint *iface_hint = node->data;
1258 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1259 /* Check if the datapath already has this port. */
1260 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1261 sset_add(&ofproto->ports, node->name);
1264 free(iface_hint->br_name);
1265 free(iface_hint->br_type);
1267 shash_delete(&init_ofp_ports, node);
1271 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1272 hash_string(ofproto->up.name, 0));
1273 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1275 ofproto_init_tables(ofproto_, N_TABLES);
1276 error = add_internal_flows(ofproto);
1277 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1283 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1284 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1286 struct ofputil_flow_mod fm;
1289 match_init_catchall(&fm.match);
1291 match_set_reg(&fm.match, 0, id);
1292 fm.new_cookie = htonll(0);
1293 fm.cookie = htonll(0);
1294 fm.cookie_mask = htonll(0);
1295 fm.table_id = TBL_INTERNAL;
1296 fm.command = OFPFC_ADD;
1297 fm.idle_timeout = 0;
1298 fm.hard_timeout = 0;
1302 fm.ofpacts = ofpacts->data;
1303 fm.ofpacts_len = ofpacts->size;
1305 error = ofproto_flow_mod(&ofproto->up, &fm);
1307 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1308 id, ofperr_to_string(error));
1312 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1313 ovs_assert(*rulep != NULL);
1319 add_internal_flows(struct ofproto_dpif *ofproto)
1321 struct ofpact_controller *controller;
1322 uint64_t ofpacts_stub[128 / 8];
1323 struct ofpbuf ofpacts;
1327 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1330 controller = ofpact_put_CONTROLLER(&ofpacts);
1331 controller->max_len = UINT16_MAX;
1332 controller->controller_id = 0;
1333 controller->reason = OFPR_NO_MATCH;
1334 ofpact_pad(&ofpacts);
1336 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1341 ofpbuf_clear(&ofpacts);
1342 error = add_internal_flow(ofproto, id++, &ofpacts,
1343 &ofproto->no_packet_in_rule);
1348 complete_operations(struct ofproto_dpif *ofproto)
1350 struct dpif_completion *c, *next;
1352 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1353 ofoperation_complete(c->op, 0);
1354 list_remove(&c->list_node);
1360 destruct(struct ofproto *ofproto_)
1362 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1363 struct rule_dpif *rule, *next_rule;
1364 struct oftable *table;
1367 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1368 complete_operations(ofproto);
1370 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1371 struct cls_cursor cursor;
1373 cls_cursor_init(&cursor, &table->cls, NULL);
1374 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1375 ofproto_rule_destroy(&rule->up);
1379 for (i = 0; i < MAX_MIRRORS; i++) {
1380 mirror_destroy(ofproto->mirrors[i]);
1383 netflow_destroy(ofproto->netflow);
1384 dpif_sflow_destroy(ofproto->sflow);
1385 hmap_destroy(&ofproto->bundles);
1386 mac_learning_destroy(ofproto->ml);
1388 hmap_destroy(&ofproto->facets);
1389 hmap_destroy(&ofproto->subfacets);
1390 governor_destroy(ofproto->governor);
1392 hmap_destroy(&ofproto->vlandev_map);
1393 hmap_destroy(&ofproto->realdev_vid_map);
1395 sset_destroy(&ofproto->ports);
1396 sset_destroy(&ofproto->ghost_ports);
1397 sset_destroy(&ofproto->port_poll_set);
1399 close_dpif_backer(ofproto->backer);
1403 run_fast(struct ofproto *ofproto_)
1405 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1406 struct ofport_dpif *ofport;
1408 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1409 port_run_fast(ofport);
1416 run(struct ofproto *ofproto_)
1418 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1419 struct ofport_dpif *ofport;
1420 struct ofbundle *bundle;
1424 complete_operations(ofproto);
1427 error = run_fast(ofproto_);
1432 if (ofproto->netflow) {
1433 if (netflow_run(ofproto->netflow)) {
1434 send_netflow_active_timeouts(ofproto);
1437 if (ofproto->sflow) {
1438 dpif_sflow_run(ofproto->sflow);
1441 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1444 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1449 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1451 /* Check the consistency of a random facet, to aid debugging. */
1452 if (!hmap_is_empty(&ofproto->facets)
1453 && !ofproto->backer->need_revalidate) {
1454 struct facet *facet;
1456 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1457 struct facet, hmap_node);
1458 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1460 if (!facet_check_consistency(facet)) {
1461 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1466 if (ofproto->governor) {
1469 governor_run(ofproto->governor);
1471 /* If the governor has shrunk to its minimum size and the number of
1472 * subfacets has dwindled, then drop the governor entirely.
1474 * For hysteresis, the number of subfacets to drop the governor is
1475 * smaller than the number needed to trigger its creation. */
1476 n_subfacets = hmap_count(&ofproto->subfacets);
1477 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1478 && governor_is_idle(ofproto->governor)) {
1479 governor_destroy(ofproto->governor);
1480 ofproto->governor = NULL;
1488 wait(struct ofproto *ofproto_)
1490 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1491 struct ofport_dpif *ofport;
1492 struct ofbundle *bundle;
1494 if (!clogged && !list_is_empty(&ofproto->completions)) {
1495 poll_immediate_wake();
1498 dpif_wait(ofproto->backer->dpif);
1499 dpif_recv_wait(ofproto->backer->dpif);
1500 if (ofproto->sflow) {
1501 dpif_sflow_wait(ofproto->sflow);
1503 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1504 poll_immediate_wake();
1506 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1509 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1510 bundle_wait(bundle);
1512 if (ofproto->netflow) {
1513 netflow_wait(ofproto->netflow);
1515 mac_learning_wait(ofproto->ml);
1517 if (ofproto->backer->need_revalidate) {
1518 /* Shouldn't happen, but if it does just go around again. */
1519 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1520 poll_immediate_wake();
1522 if (ofproto->governor) {
1523 governor_wait(ofproto->governor);
1528 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1530 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1532 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1533 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1537 flush(struct ofproto *ofproto_)
1539 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1540 struct subfacet *subfacet, *next_subfacet;
1541 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1545 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1546 &ofproto->subfacets) {
1547 if (subfacet->path != SF_NOT_INSTALLED) {
1548 batch[n_batch++] = subfacet;
1549 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1550 subfacet_destroy_batch(ofproto, batch, n_batch);
1554 subfacet_destroy(subfacet);
1559 subfacet_destroy_batch(ofproto, batch, n_batch);
1564 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1565 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1567 *arp_match_ip = true;
1568 *actions = (OFPUTIL_A_OUTPUT |
1569 OFPUTIL_A_SET_VLAN_VID |
1570 OFPUTIL_A_SET_VLAN_PCP |
1571 OFPUTIL_A_STRIP_VLAN |
1572 OFPUTIL_A_SET_DL_SRC |
1573 OFPUTIL_A_SET_DL_DST |
1574 OFPUTIL_A_SET_NW_SRC |
1575 OFPUTIL_A_SET_NW_DST |
1576 OFPUTIL_A_SET_NW_TOS |
1577 OFPUTIL_A_SET_TP_SRC |
1578 OFPUTIL_A_SET_TP_DST |
1583 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1585 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1586 struct dpif_dp_stats s;
1588 strcpy(ots->name, "classifier");
1590 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1592 ots->lookup_count = htonll(s.n_hit + s.n_missed);
1593 ots->matched_count = htonll(s.n_hit + ofproto->n_matches);
1596 static struct ofport *
1599 struct ofport_dpif *port = xmalloc(sizeof *port);
1604 port_dealloc(struct ofport *port_)
1606 struct ofport_dpif *port = ofport_dpif_cast(port_);
1611 port_construct(struct ofport *port_)
1613 struct ofport_dpif *port = ofport_dpif_cast(port_);
1614 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1615 const struct netdev *netdev = port->up.netdev;
1616 struct dpif_port dpif_port;
1619 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1620 port->bundle = NULL;
1622 port->tag = tag_create_random();
1623 port->may_enable = true;
1624 port->stp_port = NULL;
1625 port->stp_state = STP_DISABLED;
1626 port->tnl_port = NULL;
1627 hmap_init(&port->priorities);
1628 port->realdev_ofp_port = 0;
1629 port->vlandev_vid = 0;
1630 port->carrier_seq = netdev_get_carrier_resets(netdev);
1632 if (netdev_vport_is_patch(netdev)) {
1633 /* XXX By bailing out here, we don't do required sFlow work. */
1634 port->odp_port = OVSP_NONE;
1638 error = dpif_port_query_by_name(ofproto->backer->dpif,
1639 netdev_vport_get_dpif_port(netdev),
1645 port->odp_port = dpif_port.port_no;
1647 if (netdev_get_tunnel_config(netdev)) {
1648 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1650 /* Sanity-check that a mapping doesn't already exist. This
1651 * shouldn't happen for non-tunnel ports. */
1652 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1653 VLOG_ERR("port %s already has an OpenFlow port number",
1655 dpif_port_destroy(&dpif_port);
1659 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1660 hash_int(port->odp_port, 0));
1662 dpif_port_destroy(&dpif_port);
1664 if (ofproto->sflow) {
1665 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1672 port_destruct(struct ofport *port_)
1674 struct ofport_dpif *port = ofport_dpif_cast(port_);
1675 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1676 const char *dp_port_name = netdev_vport_get_dpif_port(port->up.netdev);
1677 const char *devname = netdev_get_name(port->up.netdev);
1679 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1680 /* The underlying device is still there, so delete it. This
1681 * happens when the ofproto is being destroyed, since the caller
1682 * assumes that removal of attached ports will happen as part of
1684 if (!port->tnl_port) {
1685 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1687 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1690 if (port->odp_port != OVSP_NONE && !port->tnl_port) {
1691 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1694 tnl_port_del(port->tnl_port);
1695 sset_find_and_delete(&ofproto->ports, devname);
1696 sset_find_and_delete(&ofproto->ghost_ports, devname);
1697 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1698 bundle_remove(port_);
1699 set_cfm(port_, NULL);
1700 if (ofproto->sflow) {
1701 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1704 ofport_clear_priorities(port);
1705 hmap_destroy(&port->priorities);
1709 port_modified(struct ofport *port_)
1711 struct ofport_dpif *port = ofport_dpif_cast(port_);
1713 if (port->bundle && port->bundle->bond) {
1714 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1719 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1721 struct ofport_dpif *port = ofport_dpif_cast(port_);
1722 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1723 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1725 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1726 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1727 OFPUTIL_PC_NO_PACKET_IN)) {
1728 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1730 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1731 bundle_update(port->bundle);
1737 set_sflow(struct ofproto *ofproto_,
1738 const struct ofproto_sflow_options *sflow_options)
1740 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1741 struct dpif_sflow *ds = ofproto->sflow;
1743 if (sflow_options) {
1745 struct ofport_dpif *ofport;
1747 ds = ofproto->sflow = dpif_sflow_create();
1748 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1749 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1751 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1753 dpif_sflow_set_options(ds, sflow_options);
1756 dpif_sflow_destroy(ds);
1757 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1758 ofproto->sflow = NULL;
1765 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1767 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1774 struct ofproto_dpif *ofproto;
1776 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1777 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1778 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1781 if (cfm_configure(ofport->cfm, s)) {
1787 cfm_destroy(ofport->cfm);
1793 get_cfm_fault(const struct ofport *ofport_)
1795 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1797 return ofport->cfm ? cfm_get_fault(ofport->cfm) : -1;
1801 get_cfm_opup(const struct ofport *ofport_)
1803 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1805 return ofport->cfm ? cfm_get_opup(ofport->cfm) : -1;
1809 get_cfm_remote_mpids(const struct ofport *ofport_, const uint64_t **rmps,
1812 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1815 cfm_get_remote_mpids(ofport->cfm, rmps, n_rmps);
1823 get_cfm_health(const struct ofport *ofport_)
1825 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1827 return ofport->cfm ? cfm_get_health(ofport->cfm) : -1;
1830 /* Spanning Tree. */
1833 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1835 struct ofproto_dpif *ofproto = ofproto_;
1836 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1837 struct ofport_dpif *ofport;
1839 ofport = stp_port_get_aux(sp);
1841 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1842 ofproto->up.name, port_num);
1844 struct eth_header *eth = pkt->l2;
1846 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1847 if (eth_addr_is_zero(eth->eth_src)) {
1848 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1849 "with unknown MAC", ofproto->up.name, port_num);
1851 send_packet(ofport, pkt);
1857 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1859 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1861 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1863 /* Only revalidate flows if the configuration changed. */
1864 if (!s != !ofproto->stp) {
1865 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1869 if (!ofproto->stp) {
1870 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1871 send_bpdu_cb, ofproto);
1872 ofproto->stp_last_tick = time_msec();
1875 stp_set_bridge_id(ofproto->stp, s->system_id);
1876 stp_set_bridge_priority(ofproto->stp, s->priority);
1877 stp_set_hello_time(ofproto->stp, s->hello_time);
1878 stp_set_max_age(ofproto->stp, s->max_age);
1879 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1881 struct ofport *ofport;
1883 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1884 set_stp_port(ofport, NULL);
1887 stp_destroy(ofproto->stp);
1888 ofproto->stp = NULL;
1895 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1897 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1901 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1902 s->designated_root = stp_get_designated_root(ofproto->stp);
1903 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1912 update_stp_port_state(struct ofport_dpif *ofport)
1914 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1915 enum stp_state state;
1917 /* Figure out new state. */
1918 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1922 if (ofport->stp_state != state) {
1923 enum ofputil_port_state of_state;
1926 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1927 netdev_get_name(ofport->up.netdev),
1928 stp_state_name(ofport->stp_state),
1929 stp_state_name(state));
1930 if (stp_learn_in_state(ofport->stp_state)
1931 != stp_learn_in_state(state)) {
1932 /* xxx Learning action flows should also be flushed. */
1933 mac_learning_flush(ofproto->ml,
1934 &ofproto->backer->revalidate_set);
1936 fwd_change = stp_forward_in_state(ofport->stp_state)
1937 != stp_forward_in_state(state);
1939 ofproto->backer->need_revalidate = REV_STP;
1940 ofport->stp_state = state;
1941 ofport->stp_state_entered = time_msec();
1943 if (fwd_change && ofport->bundle) {
1944 bundle_update(ofport->bundle);
1947 /* Update the STP state bits in the OpenFlow port description. */
1948 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1949 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1950 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1951 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1952 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1954 ofproto_port_set_state(&ofport->up, of_state);
1958 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1959 * caller is responsible for assigning STP port numbers and ensuring
1960 * there are no duplicates. */
1962 set_stp_port(struct ofport *ofport_,
1963 const struct ofproto_port_stp_settings *s)
1965 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1966 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1967 struct stp_port *sp = ofport->stp_port;
1969 if (!s || !s->enable) {
1971 ofport->stp_port = NULL;
1972 stp_port_disable(sp);
1973 update_stp_port_state(ofport);
1976 } else if (sp && stp_port_no(sp) != s->port_num
1977 && ofport == stp_port_get_aux(sp)) {
1978 /* The port-id changed, so disable the old one if it's not
1979 * already in use by another port. */
1980 stp_port_disable(sp);
1983 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1984 stp_port_enable(sp);
1986 stp_port_set_aux(sp, ofport);
1987 stp_port_set_priority(sp, s->priority);
1988 stp_port_set_path_cost(sp, s->path_cost);
1990 update_stp_port_state(ofport);
1996 get_stp_port_status(struct ofport *ofport_,
1997 struct ofproto_port_stp_status *s)
1999 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2000 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2001 struct stp_port *sp = ofport->stp_port;
2003 if (!ofproto->stp || !sp) {
2009 s->port_id = stp_port_get_id(sp);
2010 s->state = stp_port_get_state(sp);
2011 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2012 s->role = stp_port_get_role(sp);
2013 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2019 stp_run(struct ofproto_dpif *ofproto)
2022 long long int now = time_msec();
2023 long long int elapsed = now - ofproto->stp_last_tick;
2024 struct stp_port *sp;
2027 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2028 ofproto->stp_last_tick = now;
2030 while (stp_get_changed_port(ofproto->stp, &sp)) {
2031 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2034 update_stp_port_state(ofport);
2038 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2039 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2045 stp_wait(struct ofproto_dpif *ofproto)
2048 poll_timer_wait(1000);
2052 /* Returns true if STP should process 'flow'. */
2054 stp_should_process_flow(const struct flow *flow)
2056 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2060 stp_process_packet(const struct ofport_dpif *ofport,
2061 const struct ofpbuf *packet)
2063 struct ofpbuf payload = *packet;
2064 struct eth_header *eth = payload.data;
2065 struct stp_port *sp = ofport->stp_port;
2067 /* Sink packets on ports that have STP disabled when the bridge has
2069 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2073 /* Trim off padding on payload. */
2074 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2075 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2078 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2079 stp_received_bpdu(sp, payload.data, payload.size);
2083 static struct priority_to_dscp *
2084 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2086 struct priority_to_dscp *pdscp;
2089 hash = hash_int(priority, 0);
2090 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2091 if (pdscp->priority == priority) {
2099 ofport_clear_priorities(struct ofport_dpif *ofport)
2101 struct priority_to_dscp *pdscp, *next;
2103 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2104 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2110 set_queues(struct ofport *ofport_,
2111 const struct ofproto_port_queue *qdscp_list,
2114 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2115 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2116 struct hmap new = HMAP_INITIALIZER(&new);
2119 for (i = 0; i < n_qdscp; i++) {
2120 struct priority_to_dscp *pdscp;
2124 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2125 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2130 pdscp = get_priority(ofport, priority);
2132 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2134 pdscp = xmalloc(sizeof *pdscp);
2135 pdscp->priority = priority;
2137 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2140 if (pdscp->dscp != dscp) {
2142 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2145 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2148 if (!hmap_is_empty(&ofport->priorities)) {
2149 ofport_clear_priorities(ofport);
2150 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2153 hmap_swap(&new, &ofport->priorities);
2161 /* Expires all MAC learning entries associated with 'bundle' and forces its
2162 * ofproto to revalidate every flow.
2164 * Normally MAC learning entries are removed only from the ofproto associated
2165 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2166 * are removed from every ofproto. When patch ports and SLB bonds are in use
2167 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2168 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2169 * with the host from which it migrated. */
2171 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2173 struct ofproto_dpif *ofproto = bundle->ofproto;
2174 struct mac_learning *ml = ofproto->ml;
2175 struct mac_entry *mac, *next_mac;
2177 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2178 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2179 if (mac->port.p == bundle) {
2181 struct ofproto_dpif *o;
2183 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2185 struct mac_entry *e;
2187 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2190 mac_learning_expire(o->ml, e);
2196 mac_learning_expire(ml, mac);
2201 static struct ofbundle *
2202 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2204 struct ofbundle *bundle;
2206 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2207 &ofproto->bundles) {
2208 if (bundle->aux == aux) {
2215 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2216 * ones that are found to 'bundles'. */
2218 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2219 void **auxes, size_t n_auxes,
2220 struct hmapx *bundles)
2224 hmapx_init(bundles);
2225 for (i = 0; i < n_auxes; i++) {
2226 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2228 hmapx_add(bundles, bundle);
2234 bundle_update(struct ofbundle *bundle)
2236 struct ofport_dpif *port;
2238 bundle->floodable = true;
2239 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2240 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2241 || !stp_forward_in_state(port->stp_state)) {
2242 bundle->floodable = false;
2249 bundle_del_port(struct ofport_dpif *port)
2251 struct ofbundle *bundle = port->bundle;
2253 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2255 list_remove(&port->bundle_node);
2256 port->bundle = NULL;
2259 lacp_slave_unregister(bundle->lacp, port);
2262 bond_slave_unregister(bundle->bond, port);
2265 bundle_update(bundle);
2269 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
2270 struct lacp_slave_settings *lacp,
2271 uint32_t bond_stable_id)
2273 struct ofport_dpif *port;
2275 port = get_ofp_port(bundle->ofproto, ofp_port);
2280 if (port->bundle != bundle) {
2281 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2283 bundle_del_port(port);
2286 port->bundle = bundle;
2287 list_push_back(&bundle->ports, &port->bundle_node);
2288 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2289 || !stp_forward_in_state(port->stp_state)) {
2290 bundle->floodable = false;
2294 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2295 lacp_slave_register(bundle->lacp, port, lacp);
2298 port->bond_stable_id = bond_stable_id;
2304 bundle_destroy(struct ofbundle *bundle)
2306 struct ofproto_dpif *ofproto;
2307 struct ofport_dpif *port, *next_port;
2314 ofproto = bundle->ofproto;
2315 for (i = 0; i < MAX_MIRRORS; i++) {
2316 struct ofmirror *m = ofproto->mirrors[i];
2318 if (m->out == bundle) {
2320 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2321 || hmapx_find_and_delete(&m->dsts, bundle)) {
2322 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2327 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2328 bundle_del_port(port);
2331 bundle_flush_macs(bundle, true);
2332 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2334 free(bundle->trunks);
2335 lacp_destroy(bundle->lacp);
2336 bond_destroy(bundle->bond);
2341 bundle_set(struct ofproto *ofproto_, void *aux,
2342 const struct ofproto_bundle_settings *s)
2344 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2345 bool need_flush = false;
2346 struct ofport_dpif *port;
2347 struct ofbundle *bundle;
2348 unsigned long *trunks;
2354 bundle_destroy(bundle_lookup(ofproto, aux));
2358 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2359 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2361 bundle = bundle_lookup(ofproto, aux);
2363 bundle = xmalloc(sizeof *bundle);
2365 bundle->ofproto = ofproto;
2366 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2367 hash_pointer(aux, 0));
2369 bundle->name = NULL;
2371 list_init(&bundle->ports);
2372 bundle->vlan_mode = PORT_VLAN_TRUNK;
2374 bundle->trunks = NULL;
2375 bundle->use_priority_tags = s->use_priority_tags;
2376 bundle->lacp = NULL;
2377 bundle->bond = NULL;
2379 bundle->floodable = true;
2381 bundle->src_mirrors = 0;
2382 bundle->dst_mirrors = 0;
2383 bundle->mirror_out = 0;
2386 if (!bundle->name || strcmp(s->name, bundle->name)) {
2388 bundle->name = xstrdup(s->name);
2393 if (!bundle->lacp) {
2394 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2395 bundle->lacp = lacp_create();
2397 lacp_configure(bundle->lacp, s->lacp);
2399 lacp_destroy(bundle->lacp);
2400 bundle->lacp = NULL;
2403 /* Update set of ports. */
2405 for (i = 0; i < s->n_slaves; i++) {
2406 if (!bundle_add_port(bundle, s->slaves[i],
2407 s->lacp ? &s->lacp_slaves[i] : NULL,
2408 s->bond_stable_ids ? s->bond_stable_ids[i] : 0)) {
2412 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2413 struct ofport_dpif *next_port;
2415 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2416 for (i = 0; i < s->n_slaves; i++) {
2417 if (s->slaves[i] == port->up.ofp_port) {
2422 bundle_del_port(port);
2426 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2428 if (list_is_empty(&bundle->ports)) {
2429 bundle_destroy(bundle);
2433 /* Set VLAN tagging mode */
2434 if (s->vlan_mode != bundle->vlan_mode
2435 || s->use_priority_tags != bundle->use_priority_tags) {
2436 bundle->vlan_mode = s->vlan_mode;
2437 bundle->use_priority_tags = s->use_priority_tags;
2442 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2443 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2445 if (vlan != bundle->vlan) {
2446 bundle->vlan = vlan;
2450 /* Get trunked VLANs. */
2451 switch (s->vlan_mode) {
2452 case PORT_VLAN_ACCESS:
2456 case PORT_VLAN_TRUNK:
2457 trunks = CONST_CAST(unsigned long *, s->trunks);
2460 case PORT_VLAN_NATIVE_UNTAGGED:
2461 case PORT_VLAN_NATIVE_TAGGED:
2462 if (vlan != 0 && (!s->trunks
2463 || !bitmap_is_set(s->trunks, vlan)
2464 || bitmap_is_set(s->trunks, 0))) {
2465 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2467 trunks = bitmap_clone(s->trunks, 4096);
2469 trunks = bitmap_allocate1(4096);
2471 bitmap_set1(trunks, vlan);
2472 bitmap_set0(trunks, 0);
2474 trunks = CONST_CAST(unsigned long *, s->trunks);
2481 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2482 free(bundle->trunks);
2483 if (trunks == s->trunks) {
2484 bundle->trunks = vlan_bitmap_clone(trunks);
2486 bundle->trunks = trunks;
2491 if (trunks != s->trunks) {
2496 if (!list_is_short(&bundle->ports)) {
2497 bundle->ofproto->has_bonded_bundles = true;
2499 if (bond_reconfigure(bundle->bond, s->bond)) {
2500 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2503 bundle->bond = bond_create(s->bond);
2504 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2507 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2508 bond_slave_register(bundle->bond, port, port->bond_stable_id,
2512 bond_destroy(bundle->bond);
2513 bundle->bond = NULL;
2516 /* If we changed something that would affect MAC learning, un-learn
2517 * everything on this port and force flow revalidation. */
2519 bundle_flush_macs(bundle, false);
2526 bundle_remove(struct ofport *port_)
2528 struct ofport_dpif *port = ofport_dpif_cast(port_);
2529 struct ofbundle *bundle = port->bundle;
2532 bundle_del_port(port);
2533 if (list_is_empty(&bundle->ports)) {
2534 bundle_destroy(bundle);
2535 } else if (list_is_short(&bundle->ports)) {
2536 bond_destroy(bundle->bond);
2537 bundle->bond = NULL;
2543 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2545 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2546 struct ofport_dpif *port = port_;
2547 uint8_t ea[ETH_ADDR_LEN];
2550 error = netdev_get_etheraddr(port->up.netdev, ea);
2552 struct ofpbuf packet;
2555 ofpbuf_init(&packet, 0);
2556 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2558 memcpy(packet_pdu, pdu, pdu_size);
2560 send_packet(port, &packet);
2561 ofpbuf_uninit(&packet);
2563 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2564 "%s (%s)", port->bundle->name,
2565 netdev_get_name(port->up.netdev), strerror(error));
2570 bundle_send_learning_packets(struct ofbundle *bundle)
2572 struct ofproto_dpif *ofproto = bundle->ofproto;
2573 int error, n_packets, n_errors;
2574 struct mac_entry *e;
2576 error = n_packets = n_errors = 0;
2577 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2578 if (e->port.p != bundle) {
2579 struct ofpbuf *learning_packet;
2580 struct ofport_dpif *port;
2584 /* The assignment to "port" is unnecessary but makes "grep"ing for
2585 * struct ofport_dpif more effective. */
2586 learning_packet = bond_compose_learning_packet(bundle->bond,
2590 ret = send_packet(port, learning_packet);
2591 ofpbuf_delete(learning_packet);
2601 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2602 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2603 "packets, last error was: %s",
2604 bundle->name, n_errors, n_packets, strerror(error));
2606 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2607 bundle->name, n_packets);
2612 bundle_run(struct ofbundle *bundle)
2615 lacp_run(bundle->lacp, send_pdu_cb);
2618 struct ofport_dpif *port;
2620 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2621 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2624 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2625 lacp_status(bundle->lacp));
2626 if (bond_should_send_learning_packets(bundle->bond)) {
2627 bundle_send_learning_packets(bundle);
2633 bundle_wait(struct ofbundle *bundle)
2636 lacp_wait(bundle->lacp);
2639 bond_wait(bundle->bond);
2646 mirror_scan(struct ofproto_dpif *ofproto)
2650 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2651 if (!ofproto->mirrors[idx]) {
2658 static struct ofmirror *
2659 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2663 for (i = 0; i < MAX_MIRRORS; i++) {
2664 struct ofmirror *mirror = ofproto->mirrors[i];
2665 if (mirror && mirror->aux == aux) {
2673 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2675 mirror_update_dups(struct ofproto_dpif *ofproto)
2679 for (i = 0; i < MAX_MIRRORS; i++) {
2680 struct ofmirror *m = ofproto->mirrors[i];
2683 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2687 for (i = 0; i < MAX_MIRRORS; i++) {
2688 struct ofmirror *m1 = ofproto->mirrors[i];
2695 for (j = i + 1; j < MAX_MIRRORS; j++) {
2696 struct ofmirror *m2 = ofproto->mirrors[j];
2698 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2699 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2700 m2->dup_mirrors |= m1->dup_mirrors;
2707 mirror_set(struct ofproto *ofproto_, void *aux,
2708 const struct ofproto_mirror_settings *s)
2710 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2711 mirror_mask_t mirror_bit;
2712 struct ofbundle *bundle;
2713 struct ofmirror *mirror;
2714 struct ofbundle *out;
2715 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2716 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2719 mirror = mirror_lookup(ofproto, aux);
2721 mirror_destroy(mirror);
2727 idx = mirror_scan(ofproto);
2729 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2731 ofproto->up.name, MAX_MIRRORS, s->name);
2735 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2736 mirror->ofproto = ofproto;
2739 mirror->out_vlan = -1;
2740 mirror->name = NULL;
2743 if (!mirror->name || strcmp(s->name, mirror->name)) {
2745 mirror->name = xstrdup(s->name);
2748 /* Get the new configuration. */
2749 if (s->out_bundle) {
2750 out = bundle_lookup(ofproto, s->out_bundle);
2752 mirror_destroy(mirror);
2758 out_vlan = s->out_vlan;
2760 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2761 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2763 /* If the configuration has not changed, do nothing. */
2764 if (hmapx_equals(&srcs, &mirror->srcs)
2765 && hmapx_equals(&dsts, &mirror->dsts)
2766 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2767 && mirror->out == out
2768 && mirror->out_vlan == out_vlan)
2770 hmapx_destroy(&srcs);
2771 hmapx_destroy(&dsts);
2775 hmapx_swap(&srcs, &mirror->srcs);
2776 hmapx_destroy(&srcs);
2778 hmapx_swap(&dsts, &mirror->dsts);
2779 hmapx_destroy(&dsts);
2781 free(mirror->vlans);
2782 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2785 mirror->out_vlan = out_vlan;
2787 /* Update bundles. */
2788 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2789 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2790 if (hmapx_contains(&mirror->srcs, bundle)) {
2791 bundle->src_mirrors |= mirror_bit;
2793 bundle->src_mirrors &= ~mirror_bit;
2796 if (hmapx_contains(&mirror->dsts, bundle)) {
2797 bundle->dst_mirrors |= mirror_bit;
2799 bundle->dst_mirrors &= ~mirror_bit;
2802 if (mirror->out == bundle) {
2803 bundle->mirror_out |= mirror_bit;
2805 bundle->mirror_out &= ~mirror_bit;
2809 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2810 ofproto->has_mirrors = true;
2811 mac_learning_flush(ofproto->ml,
2812 &ofproto->backer->revalidate_set);
2813 mirror_update_dups(ofproto);
2819 mirror_destroy(struct ofmirror *mirror)
2821 struct ofproto_dpif *ofproto;
2822 mirror_mask_t mirror_bit;
2823 struct ofbundle *bundle;
2830 ofproto = mirror->ofproto;
2831 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2832 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2834 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2835 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2836 bundle->src_mirrors &= ~mirror_bit;
2837 bundle->dst_mirrors &= ~mirror_bit;
2838 bundle->mirror_out &= ~mirror_bit;
2841 hmapx_destroy(&mirror->srcs);
2842 hmapx_destroy(&mirror->dsts);
2843 free(mirror->vlans);
2845 ofproto->mirrors[mirror->idx] = NULL;
2849 mirror_update_dups(ofproto);
2851 ofproto->has_mirrors = false;
2852 for (i = 0; i < MAX_MIRRORS; i++) {
2853 if (ofproto->mirrors[i]) {
2854 ofproto->has_mirrors = true;
2861 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2862 uint64_t *packets, uint64_t *bytes)
2864 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2865 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2868 *packets = *bytes = UINT64_MAX;
2872 *packets = mirror->packet_count;
2873 *bytes = mirror->byte_count;
2879 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2881 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2882 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2883 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2889 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2891 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2892 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2893 return bundle && bundle->mirror_out != 0;
2897 forward_bpdu_changed(struct ofproto *ofproto_)
2899 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2900 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2904 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2907 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2908 mac_learning_set_idle_time(ofproto->ml, idle_time);
2909 mac_learning_set_max_entries(ofproto->ml, max_entries);
2914 static struct ofport_dpif *
2915 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
2917 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2918 return ofport ? ofport_dpif_cast(ofport) : NULL;
2921 static struct ofport_dpif *
2922 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
2924 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2925 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2929 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2930 struct ofproto_port *ofproto_port,
2931 struct dpif_port *dpif_port)
2933 ofproto_port->name = dpif_port->name;
2934 ofproto_port->type = dpif_port->type;
2935 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2938 static struct ofport_dpif *
2939 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
2941 const struct ofproto_dpif *ofproto;
2944 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
2949 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2950 struct ofport *ofport;
2952 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
2953 if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) {
2954 return ofport_dpif_cast(ofport);
2961 port_run_fast(struct ofport_dpif *ofport)
2963 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2964 struct ofpbuf packet;
2966 ofpbuf_init(&packet, 0);
2967 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2968 send_packet(ofport, &packet);
2969 ofpbuf_uninit(&packet);
2974 port_run(struct ofport_dpif *ofport)
2976 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2977 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2978 bool enable = netdev_get_carrier(ofport->up.netdev);
2980 ofport->carrier_seq = carrier_seq;
2982 port_run_fast(ofport);
2984 if (ofport->tnl_port
2985 && tnl_port_reconfigure(&ofport->up, ofport->odp_port,
2986 &ofport->tnl_port)) {
2987 ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true;
2991 int cfm_opup = cfm_get_opup(ofport->cfm);
2993 cfm_run(ofport->cfm);
2994 enable = enable && !cfm_get_fault(ofport->cfm);
2996 if (cfm_opup >= 0) {
2997 enable = enable && cfm_opup;
3001 if (ofport->bundle) {
3002 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3003 if (carrier_changed) {
3004 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3008 if (ofport->may_enable != enable) {
3009 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3011 if (ofproto->has_bundle_action) {
3012 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3016 ofport->may_enable = enable;
3020 port_wait(struct ofport_dpif *ofport)
3023 cfm_wait(ofport->cfm);
3028 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3029 struct ofproto_port *ofproto_port)
3031 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3032 struct dpif_port dpif_port;
3035 if (sset_contains(&ofproto->ghost_ports, devname)) {
3036 const char *type = netdev_get_type_from_name(devname);
3038 /* We may be called before ofproto->up.port_by_name is populated with
3039 * the appropriate ofport. For this reason, we must get the name and
3040 * type from the netdev layer directly. */
3042 const struct ofport *ofport;
3044 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3045 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3046 ofproto_port->name = xstrdup(devname);
3047 ofproto_port->type = xstrdup(type);
3053 if (!sset_contains(&ofproto->ports, devname)) {
3056 error = dpif_port_query_by_name(ofproto->backer->dpif,
3057 devname, &dpif_port);
3059 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3065 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3067 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3068 const char *dp_port_name = netdev_vport_get_dpif_port(netdev);
3069 const char *devname = netdev_get_name(netdev);
3071 if (netdev_vport_is_patch(netdev)) {
3072 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3076 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3077 uint32_t port_no = UINT32_MAX;
3080 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3084 if (netdev_get_tunnel_config(netdev)) {
3085 simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no);
3089 if (netdev_get_tunnel_config(netdev)) {
3090 sset_add(&ofproto->ghost_ports, devname);
3092 sset_add(&ofproto->ports, devname);
3098 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
3100 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3101 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3108 sset_find_and_delete(&ofproto->ghost_ports,
3109 netdev_get_name(ofport->up.netdev));
3110 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3111 if (!ofport->tnl_port) {
3112 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3114 /* The caller is going to close ofport->up.netdev. If this is a
3115 * bonded port, then the bond is using that netdev, so remove it
3116 * from the bond. The client will need to reconfigure everything
3117 * after deleting ports, so then the slave will get re-added. */
3118 bundle_remove(&ofport->up);
3125 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3127 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3130 error = netdev_get_stats(ofport->up.netdev, stats);
3132 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3133 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3135 /* ofproto->stats.tx_packets represents packets that we created
3136 * internally and sent to some port (e.g. packets sent with
3137 * send_packet()). Account for them as if they had come from
3138 * OFPP_LOCAL and got forwarded. */
3140 if (stats->rx_packets != UINT64_MAX) {
3141 stats->rx_packets += ofproto->stats.tx_packets;
3144 if (stats->rx_bytes != UINT64_MAX) {
3145 stats->rx_bytes += ofproto->stats.tx_bytes;
3148 /* ofproto->stats.rx_packets represents packets that were received on
3149 * some port and we processed internally and dropped (e.g. STP).
3150 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3152 if (stats->tx_packets != UINT64_MAX) {
3153 stats->tx_packets += ofproto->stats.rx_packets;
3156 if (stats->tx_bytes != UINT64_MAX) {
3157 stats->tx_bytes += ofproto->stats.rx_bytes;
3164 /* Account packets for LOCAL port. */
3166 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
3167 size_t tx_size, size_t rx_size)
3169 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3172 ofproto->stats.rx_packets++;
3173 ofproto->stats.rx_bytes += rx_size;
3176 ofproto->stats.tx_packets++;
3177 ofproto->stats.tx_bytes += tx_size;
3181 struct port_dump_state {
3186 struct ofproto_port port;
3191 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3193 *statep = xzalloc(sizeof(struct port_dump_state));
3198 port_dump_next(const struct ofproto *ofproto_, void *state_,
3199 struct ofproto_port *port)
3201 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3202 struct port_dump_state *state = state_;
3203 const struct sset *sset;
3204 struct sset_node *node;
3206 if (state->has_port) {
3207 ofproto_port_destroy(&state->port);
3208 state->has_port = false;
3210 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3211 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3214 error = port_query_by_name(ofproto_, node->name, &state->port);
3216 *port = state->port;
3217 state->has_port = true;
3219 } else if (error != ENODEV) {
3224 if (!state->ghost) {
3225 state->ghost = true;
3228 return port_dump_next(ofproto_, state_, port);
3235 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3237 struct port_dump_state *state = state_;
3239 if (state->has_port) {
3240 ofproto_port_destroy(&state->port);
3247 port_poll(const struct ofproto *ofproto_, char **devnamep)
3249 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3251 if (ofproto->port_poll_errno) {
3252 int error = ofproto->port_poll_errno;
3253 ofproto->port_poll_errno = 0;
3257 if (sset_is_empty(&ofproto->port_poll_set)) {
3261 *devnamep = sset_pop(&ofproto->port_poll_set);
3266 port_poll_wait(const struct ofproto *ofproto_)
3268 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3269 dpif_port_poll_wait(ofproto->backer->dpif);
3273 port_is_lacp_current(const struct ofport *ofport_)
3275 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3276 return (ofport->bundle && ofport->bundle->lacp
3277 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3281 /* Upcall handling. */
3283 /* Flow miss batching.
3285 * Some dpifs implement operations faster when you hand them off in a batch.
3286 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3287 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3288 * more packets, plus possibly installing the flow in the dpif.
3290 * So far we only batch the operations that affect flow setup time the most.
3291 * It's possible to batch more than that, but the benefit might be minimal. */
3293 struct hmap_node hmap_node;
3294 struct ofproto_dpif *ofproto;
3296 enum odp_key_fitness key_fitness;
3297 const struct nlattr *key;
3299 ovs_be16 initial_tci;
3300 struct list packets;
3301 enum dpif_upcall_type upcall_type;
3302 uint32_t odp_in_port;
3305 struct flow_miss_op {
3306 struct dpif_op dpif_op;
3307 struct subfacet *subfacet; /* Subfacet */
3308 void *garbage; /* Pointer to pass to free(), NULL if none. */
3309 uint64_t stub[1024 / 8]; /* Temporary buffer. */
3312 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3313 * OpenFlow controller as necessary according to their individual
3314 * configurations. */
3316 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3317 const struct flow *flow)
3319 struct ofputil_packet_in pin;
3321 pin.packet = packet->data;
3322 pin.packet_len = packet->size;
3323 pin.reason = OFPR_NO_MATCH;
3324 pin.controller_id = 0;
3329 pin.send_len = 0; /* not used for flow table misses */
3331 flow_get_metadata(flow, &pin.fmd);
3333 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3336 static enum slow_path_reason
3337 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3338 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
3342 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3344 cfm_process_heartbeat(ofport->cfm, packet);
3347 } else if (ofport->bundle && ofport->bundle->lacp
3348 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3350 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3353 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3355 stp_process_packet(ofport, packet);
3363 static struct flow_miss *
3364 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3365 const struct flow *flow, uint32_t hash)
3367 struct flow_miss *miss;
3369 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3370 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3378 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3379 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3380 * 'miss' is associated with a subfacet the caller must also initialize the
3381 * returned op->subfacet, and if anything needs to be freed after processing
3382 * the op, the caller must initialize op->garbage also. */
3384 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3385 struct flow_miss_op *op)
3387 if (miss->flow.vlan_tci != miss->initial_tci) {
3388 /* This packet was received on a VLAN splinter port. We
3389 * added a VLAN to the packet to make the packet resemble
3390 * the flow, but the actions were composed assuming that
3391 * the packet contained no VLAN. So, we must remove the
3392 * VLAN header from the packet before trying to execute the
3394 eth_pop_vlan(packet);
3397 op->subfacet = NULL;
3399 op->dpif_op.type = DPIF_OP_EXECUTE;
3400 op->dpif_op.u.execute.key = miss->key;
3401 op->dpif_op.u.execute.key_len = miss->key_len;
3402 op->dpif_op.u.execute.packet = packet;
3405 /* Helper for handle_flow_miss_without_facet() and
3406 * handle_flow_miss_with_facet(). */
3408 handle_flow_miss_common(struct rule_dpif *rule,
3409 struct ofpbuf *packet, const struct flow *flow)
3411 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3413 ofproto->n_matches++;
3415 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3417 * Extra-special case for fail-open mode.
3419 * We are in fail-open mode and the packet matched the fail-open
3420 * rule, but we are connected to a controller too. We should send
3421 * the packet up to the controller in the hope that it will try to
3422 * set up a flow and thereby allow us to exit fail-open.
3424 * See the top-level comment in fail-open.c for more information.
3426 send_packet_in_miss(ofproto, packet, flow);
3430 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3431 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3432 * installing a datapath flow. The answer is usually "yes" (a return value of
3433 * true). However, for short flows the cost of bookkeeping is much higher than
3434 * the benefits, so when the datapath holds a large number of flows we impose
3435 * some heuristics to decide which flows are likely to be worth tracking. */
3437 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3438 struct flow_miss *miss, uint32_t hash)
3440 if (!ofproto->governor) {
3443 n_subfacets = hmap_count(&ofproto->subfacets);
3444 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3448 ofproto->governor = governor_create(ofproto->up.name);
3451 return governor_should_install_flow(ofproto->governor, hash,
3452 list_size(&miss->packets));
3455 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3456 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3457 * increment '*n_ops'. */
3459 handle_flow_miss_without_facet(struct flow_miss *miss,
3460 struct rule_dpif *rule,
3461 struct flow_miss_op *ops, size_t *n_ops)
3463 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3464 long long int now = time_msec();
3465 struct action_xlate_ctx ctx;
3466 struct ofpbuf *packet;
3468 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3469 struct flow_miss_op *op = &ops[*n_ops];
3470 struct dpif_flow_stats stats;
3471 struct ofpbuf odp_actions;
3473 COVERAGE_INC(facet_suppress);
3475 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3477 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3478 rule_credit_stats(rule, &stats);
3480 action_xlate_ctx_init(&ctx, ofproto, &miss->flow, miss->initial_tci,
3482 ctx.resubmit_stats = &stats;
3483 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
3486 if (odp_actions.size) {
3487 struct dpif_execute *execute = &op->dpif_op.u.execute;
3489 init_flow_miss_execute_op(miss, packet, op);
3490 execute->actions = odp_actions.data;
3491 execute->actions_len = odp_actions.size;
3492 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3496 ofpbuf_uninit(&odp_actions);
3501 /* Handles 'miss', which matches 'facet'. May add any required datapath
3502 * operations to 'ops', incrementing '*n_ops' for each new op.
3504 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3505 * This is really important only for new facets: if we just called time_msec()
3506 * here, then the new subfacet or its packets could look (occasionally) as
3507 * though it was used some time after the facet was used. That can make a
3508 * one-packet flow look like it has a nonzero duration, which looks odd in
3509 * e.g. NetFlow statistics. */
3511 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3513 struct flow_miss_op *ops, size_t *n_ops)
3515 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3516 enum subfacet_path want_path;
3517 struct subfacet *subfacet;
3518 struct ofpbuf *packet;
3520 subfacet = subfacet_create(facet, miss, now);
3522 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3523 struct flow_miss_op *op = &ops[*n_ops];
3524 struct dpif_flow_stats stats;
3525 struct ofpbuf odp_actions;
3527 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3529 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3530 if (!subfacet->actions || subfacet->slow) {
3531 subfacet_make_actions(subfacet, packet, &odp_actions);
3534 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3535 subfacet_update_stats(subfacet, &stats);
3537 if (subfacet->actions_len) {
3538 struct dpif_execute *execute = &op->dpif_op.u.execute;
3540 init_flow_miss_execute_op(miss, packet, op);
3541 op->subfacet = subfacet;
3542 if (!subfacet->slow) {
3543 execute->actions = subfacet->actions;
3544 execute->actions_len = subfacet->actions_len;
3545 ofpbuf_uninit(&odp_actions);
3547 execute->actions = odp_actions.data;
3548 execute->actions_len = odp_actions.size;
3549 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3554 ofpbuf_uninit(&odp_actions);
3558 want_path = subfacet_want_path(subfacet->slow);
3559 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3560 struct flow_miss_op *op = &ops[(*n_ops)++];
3561 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3563 op->subfacet = subfacet;
3565 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3566 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3567 put->key = miss->key;
3568 put->key_len = miss->key_len;
3569 if (want_path == SF_FAST_PATH) {
3570 put->actions = subfacet->actions;
3571 put->actions_len = subfacet->actions_len;
3573 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
3574 op->stub, sizeof op->stub,
3575 &put->actions, &put->actions_len);
3581 /* Handles flow miss 'miss'. May add any required datapath operations
3582 * to 'ops', incrementing '*n_ops' for each new op. */
3584 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3587 struct ofproto_dpif *ofproto = miss->ofproto;
3588 struct facet *facet;
3592 /* The caller must ensure that miss->hmap_node.hash contains
3593 * flow_hash(miss->flow, 0). */
3594 hash = miss->hmap_node.hash;
3596 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3598 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
3600 if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
3601 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
3605 facet = facet_create(rule, &miss->flow, hash);
3610 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3613 static struct drop_key *
3614 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3617 struct drop_key *drop_key;
3619 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3620 &backer->drop_keys) {
3621 if (drop_key->key_len == key_len
3622 && !memcmp(drop_key->key, key, key_len)) {
3630 drop_key_clear(struct dpif_backer *backer)
3632 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3633 struct drop_key *drop_key, *next;
3635 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3638 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3640 if (error && !VLOG_DROP_WARN(&rl)) {
3641 struct ds ds = DS_EMPTY_INITIALIZER;
3642 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3643 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3648 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3649 free(drop_key->key);
3654 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3655 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3656 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3657 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3658 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3659 * 'packet' ingressed.
3661 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3662 * 'flow''s in_port to OFPP_NONE.
3664 * This function does post-processing on data returned from
3665 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3666 * of the upcall processing logic. In particular, if the extracted in_port is
3667 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3668 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3669 * a VLAN header onto 'packet' (if it is nonnull).
3671 * Optionally, if nonnull, sets '*initial_tci' to the VLAN TCI with which the
3672 * packet was really received, that is, the actual VLAN TCI extracted by
3673 * odp_flow_key_to_flow(). (This differs from the value returned in
3674 * flow->vlan_tci only for packets received on VLAN splinters.)
3676 * Similarly, this function also includes some logic to help with tunnels. It
3677 * may modify 'flow' as necessary to make the tunneling implementation
3678 * transparent to the upcall processing logic.
3680 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3681 * or some other positive errno if there are other problems. */
3683 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3684 const struct nlattr *key, size_t key_len,
3685 struct flow *flow, enum odp_key_fitness *fitnessp,
3686 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3687 ovs_be16 *initial_tci)
3689 const struct ofport_dpif *port;
3690 enum odp_key_fitness fitness;
3693 fitness = odp_flow_key_to_flow(key, key_len, flow);
3694 if (fitness == ODP_FIT_ERROR) {
3700 *initial_tci = flow->vlan_tci;
3704 *odp_in_port = flow->in_port;
3707 if (tnl_port_should_receive(flow)) {
3708 const struct ofport *ofport = tnl_port_receive(flow);
3710 flow->in_port = OFPP_NONE;
3713 port = ofport_dpif_cast(ofport);
3715 /* We can't reproduce 'key' from 'flow'. */
3716 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3718 /* XXX: Since the tunnel module is not scoped per backer, it's
3719 * theoretically possible that we'll receive an ofport belonging to an
3720 * entirely different datapath. In practice, this can't happen because
3721 * no platforms has two separate datapaths which each support
3723 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3725 port = odp_port_to_ofport(backer, flow->in_port);
3727 flow->in_port = OFPP_NONE;
3731 flow->in_port = port->up.ofp_port;
3732 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3734 /* Make the packet resemble the flow, so that it gets sent to
3735 * an OpenFlow controller properly, so that it looks correct
3736 * for sFlow, and so that flow_extract() will get the correct
3737 * vlan_tci if it is called on 'packet'.
3739 * The allocated space inside 'packet' probably also contains
3740 * 'key', that is, both 'packet' and 'key' are probably part of
3741 * a struct dpif_upcall (see the large comment on that
3742 * structure definition), so pushing data on 'packet' is in
3743 * general not a good idea since it could overwrite 'key' or
3744 * free it as a side effect. However, it's OK in this special
3745 * case because we know that 'packet' is inside a Netlink
3746 * attribute: pushing 4 bytes will just overwrite the 4-byte
3747 * "struct nlattr", which is fine since we don't need that
3748 * header anymore. */
3749 eth_push_vlan(packet, flow->vlan_tci);
3751 /* We can't reproduce 'key' from 'flow'. */
3752 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3758 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3763 *fitnessp = fitness;
3769 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3772 struct dpif_upcall *upcall;
3773 struct flow_miss *miss;
3774 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3775 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3776 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3786 /* Construct the to-do list.
3788 * This just amounts to extracting the flow from each packet and sticking
3789 * the packets that have the same flow in the same "flow_miss" structure so
3790 * that we can process them together. */
3793 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3794 struct flow_miss *miss = &misses[n_misses];
3795 struct flow_miss *existing_miss;
3796 struct ofproto_dpif *ofproto;
3797 uint32_t odp_in_port;
3802 error = ofproto_receive(backer, upcall->packet, upcall->key,
3803 upcall->key_len, &flow, &miss->key_fitness,
3804 &ofproto, &odp_in_port, &miss->initial_tci);
3805 if (error == ENODEV) {
3806 struct drop_key *drop_key;
3808 /* Received packet on port for which we couldn't associate
3809 * an ofproto. This can happen if a port is removed while
3810 * traffic is being received. Print a rate-limited message
3811 * in case it happens frequently. Install a drop flow so
3812 * that future packets of the flow are inexpensively dropped
3814 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
3817 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
3819 drop_key = xmalloc(sizeof *drop_key);
3820 drop_key->key = xmemdup(upcall->key, upcall->key_len);
3821 drop_key->key_len = upcall->key_len;
3823 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3824 hash_bytes(drop_key->key, drop_key->key_len, 0));
3825 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3826 drop_key->key, drop_key->key_len, NULL, 0, NULL);
3833 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3834 &flow.tunnel, flow.in_port, &miss->flow);
3836 /* Add other packets to a to-do list. */
3837 hash = flow_hash(&miss->flow, 0);
3838 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
3839 if (!existing_miss) {
3840 hmap_insert(&todo, &miss->hmap_node, hash);
3841 miss->ofproto = ofproto;
3842 miss->key = upcall->key;
3843 miss->key_len = upcall->key_len;
3844 miss->upcall_type = upcall->type;
3845 miss->odp_in_port = odp_in_port;
3846 list_init(&miss->packets);
3850 miss = existing_miss;
3852 list_push_back(&miss->packets, &upcall->packet->list_node);
3855 /* Process each element in the to-do list, constructing the set of
3856 * operations to batch. */
3858 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3859 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3861 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3863 /* Execute batch. */
3864 for (i = 0; i < n_ops; i++) {
3865 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3867 dpif_operate(backer->dpif, dpif_ops, n_ops);
3869 /* Free memory and update facets. */
3870 for (i = 0; i < n_ops; i++) {
3871 struct flow_miss_op *op = &flow_miss_ops[i];
3873 switch (op->dpif_op.type) {
3874 case DPIF_OP_EXECUTE:
3877 case DPIF_OP_FLOW_PUT:
3878 if (!op->dpif_op.error) {
3879 op->subfacet->path = subfacet_want_path(op->subfacet->slow);
3883 case DPIF_OP_FLOW_DEL:
3889 hmap_destroy(&todo);
3892 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL }
3893 classify_upcall(const struct dpif_upcall *upcall)
3895 union user_action_cookie cookie;
3897 /* First look at the upcall type. */
3898 switch (upcall->type) {
3899 case DPIF_UC_ACTION:
3905 case DPIF_N_UC_TYPES:
3907 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3911 /* "action" upcalls need a closer look. */
3912 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3913 switch (cookie.type) {
3914 case USER_ACTION_COOKIE_SFLOW:
3915 return SFLOW_UPCALL;
3917 case USER_ACTION_COOKIE_SLOW_PATH:
3920 case USER_ACTION_COOKIE_UNSPEC:
3922 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64, upcall->userdata);
3928 handle_sflow_upcall(struct dpif_backer *backer,
3929 const struct dpif_upcall *upcall)
3931 struct ofproto_dpif *ofproto;
3932 union user_action_cookie cookie;
3934 uint32_t odp_in_port;
3936 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3937 &flow, NULL, &ofproto, &odp_in_port, NULL)
3938 || !ofproto->sflow) {
3942 memcpy(&cookie, &upcall->userdata, sizeof(cookie));
3943 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3944 odp_in_port, &cookie);
3948 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
3950 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3951 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3952 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3957 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
3960 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3961 struct dpif_upcall *upcall = &misses[n_misses];
3962 struct ofpbuf *buf = &miss_bufs[n_misses];
3965 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3966 sizeof miss_buf_stubs[n_misses]);
3967 error = dpif_recv(backer->dpif, upcall, buf);
3973 switch (classify_upcall(upcall)) {
3975 /* Handle it later. */
3980 handle_sflow_upcall(backer, upcall);
3990 /* Handle deferred MISS_UPCALL processing. */
3991 handle_miss_upcalls(backer, misses, n_misses);
3992 for (i = 0; i < n_misses; i++) {
3993 ofpbuf_uninit(&miss_bufs[i]);
3999 /* Flow expiration. */
4001 static int subfacet_max_idle(const struct ofproto_dpif *);
4002 static void update_stats(struct dpif_backer *);
4003 static void rule_expire(struct rule_dpif *);
4004 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
4006 /* This function is called periodically by run(). Its job is to collect
4007 * updates for the flows that have been installed into the datapath, most
4008 * importantly when they last were used, and then use that information to
4009 * expire flows that have not been used recently.
4011 * Returns the number of milliseconds after which it should be called again. */
4013 expire(struct dpif_backer *backer)
4015 struct ofproto_dpif *ofproto;
4016 int max_idle = INT32_MAX;
4018 /* Periodically clear out the drop keys in an effort to keep them
4019 * relatively few. */
4020 drop_key_clear(backer);
4022 /* Update stats for each flow in the backer. */
4023 update_stats(backer);
4025 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4026 struct rule *rule, *next_rule;
4029 if (ofproto->backer != backer) {
4033 /* Expire subfacets that have been idle too long. */
4034 dp_max_idle = subfacet_max_idle(ofproto);
4035 expire_subfacets(ofproto, dp_max_idle);
4037 max_idle = MIN(max_idle, dp_max_idle);
4039 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4041 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4042 &ofproto->up.expirable) {
4043 rule_expire(rule_dpif_cast(rule));
4046 /* All outstanding data in existing flows has been accounted, so it's a
4047 * good time to do bond rebalancing. */
4048 if (ofproto->has_bonded_bundles) {
4049 struct ofbundle *bundle;
4051 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4053 bond_rebalance(bundle->bond, &backer->revalidate_set);
4059 return MIN(max_idle, 1000);
4062 /* Updates flow table statistics given that the datapath just reported 'stats'
4063 * as 'subfacet''s statistics. */
4065 update_subfacet_stats(struct subfacet *subfacet,
4066 const struct dpif_flow_stats *stats)
4068 struct facet *facet = subfacet->facet;
4070 if (stats->n_packets >= subfacet->dp_packet_count) {
4071 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
4072 facet->packet_count += extra;
4074 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4077 if (stats->n_bytes >= subfacet->dp_byte_count) {
4078 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
4080 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4083 subfacet->dp_packet_count = stats->n_packets;
4084 subfacet->dp_byte_count = stats->n_bytes;
4086 facet->tcp_flags |= stats->tcp_flags;
4088 subfacet_update_time(subfacet, stats->used);
4089 if (facet->accounted_bytes < facet->byte_count) {
4091 facet_account(facet);
4092 facet->accounted_bytes = facet->byte_count;
4094 facet_push_stats(facet);
4097 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4098 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4100 delete_unexpected_flow(struct ofproto_dpif *ofproto,
4101 const struct nlattr *key, size_t key_len)
4103 if (!VLOG_DROP_WARN(&rl)) {
4107 odp_flow_key_format(key, key_len, &s);
4108 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
4112 COVERAGE_INC(facet_unexpected);
4113 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
4116 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4118 * This function also pushes statistics updates to rules which each facet
4119 * resubmits into. Generally these statistics will be accurate. However, if a
4120 * facet changes the rule it resubmits into at some time in between
4121 * update_stats() runs, it is possible that statistics accrued to the
4122 * old rule will be incorrectly attributed to the new rule. This could be
4123 * avoided by calling update_stats() whenever rules are created or
4124 * deleted. However, the performance impact of making so many calls to the
4125 * datapath do not justify the benefit of having perfectly accurate statistics.
4128 update_stats(struct dpif_backer *backer)
4130 const struct dpif_flow_stats *stats;
4131 struct dpif_flow_dump dump;
4132 const struct nlattr *key;
4135 dpif_flow_dump_start(&dump, backer->dpif);
4136 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4138 struct subfacet *subfacet;
4139 struct ofproto_dpif *ofproto;
4140 struct ofport_dpif *ofport;
4143 if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto,
4148 ofport = get_ofp_port(ofproto, flow.in_port);
4149 if (ofport && ofport->tnl_port) {
4150 netdev_vport_inc_rx(ofport->up.netdev, stats);
4153 key_hash = odp_flow_key_hash(key, key_len);
4154 subfacet = subfacet_find(ofproto, key, key_len, key_hash, &flow);
4155 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4157 update_subfacet_stats(subfacet, stats);
4161 /* Stats are updated per-packet. */
4164 case SF_NOT_INSTALLED:
4166 delete_unexpected_flow(ofproto, key, key_len);
4170 dpif_flow_dump_done(&dump);
4173 /* Calculates and returns the number of milliseconds of idle time after which
4174 * subfacets should expire from the datapath. When a subfacet expires, we fold
4175 * its statistics into its facet, and when a facet's last subfacet expires, we
4176 * fold its statistic into its rule. */
4178 subfacet_max_idle(const struct ofproto_dpif *ofproto)
4181 * Idle time histogram.
4183 * Most of the time a switch has a relatively small number of subfacets.
4184 * When this is the case we might as well keep statistics for all of them
4185 * in userspace and to cache them in the kernel datapath for performance as
4188 * As the number of subfacets increases, the memory required to maintain
4189 * statistics about them in userspace and in the kernel becomes
4190 * significant. However, with a large number of subfacets it is likely
4191 * that only a few of them are "heavy hitters" that consume a large amount
4192 * of bandwidth. At this point, only heavy hitters are worth caching in
4193 * the kernel and maintaining in userspaces; other subfacets we can
4196 * The technique used to compute the idle time is to build a histogram with
4197 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4198 * that is installed in the kernel gets dropped in the appropriate bucket.
4199 * After the histogram has been built, we compute the cutoff so that only
4200 * the most-recently-used 1% of subfacets (but at least
4201 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
4202 * the most-recently-used bucket of subfacets is kept, so actually an
4203 * arbitrary number of subfacets can be kept in any given expiration run
4204 * (though the next run will delete most of those unless they receive
4207 * This requires a second pass through the subfacets, in addition to the
4208 * pass made by update_stats(), because the former function never looks at
4209 * uninstallable subfacets.
4211 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4212 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4213 int buckets[N_BUCKETS] = { 0 };
4214 int total, subtotal, bucket;
4215 struct subfacet *subfacet;
4219 total = hmap_count(&ofproto->subfacets);
4220 if (total <= ofproto->up.flow_eviction_threshold) {
4221 return N_BUCKETS * BUCKET_WIDTH;
4224 /* Build histogram. */
4226 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
4227 long long int idle = now - subfacet->used;
4228 int bucket = (idle <= 0 ? 0
4229 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4230 : (unsigned int) idle / BUCKET_WIDTH);
4234 /* Find the first bucket whose flows should be expired. */
4235 subtotal = bucket = 0;
4237 subtotal += buckets[bucket++];
4238 } while (bucket < N_BUCKETS &&
4239 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
4241 if (VLOG_IS_DBG_ENABLED()) {
4245 ds_put_cstr(&s, "keep");
4246 for (i = 0; i < N_BUCKETS; i++) {
4248 ds_put_cstr(&s, ", drop");
4251 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4254 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
4258 return bucket * BUCKET_WIDTH;
4262 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
4264 /* Cutoff time for most flows. */
4265 long long int normal_cutoff = time_msec() - dp_max_idle;
4267 /* We really want to keep flows for special protocols around, so use a more
4268 * conservative cutoff. */
4269 long long int special_cutoff = time_msec() - 10000;
4271 struct subfacet *subfacet, *next_subfacet;
4272 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4276 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4277 &ofproto->subfacets) {
4278 long long int cutoff;
4280 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)
4283 if (subfacet->used < cutoff) {
4284 if (subfacet->path != SF_NOT_INSTALLED) {
4285 batch[n_batch++] = subfacet;
4286 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4287 subfacet_destroy_batch(ofproto, batch, n_batch);
4291 subfacet_destroy(subfacet);
4297 subfacet_destroy_batch(ofproto, batch, n_batch);
4301 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4302 * then delete it entirely. */
4304 rule_expire(struct rule_dpif *rule)
4306 struct facet *facet, *next_facet;
4310 if (rule->up.pending) {
4311 /* We'll have to expire it later. */
4315 /* Has 'rule' expired? */
4317 if (rule->up.hard_timeout
4318 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4319 reason = OFPRR_HARD_TIMEOUT;
4320 } else if (rule->up.idle_timeout
4321 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4322 reason = OFPRR_IDLE_TIMEOUT;
4327 COVERAGE_INC(ofproto_dpif_expired);
4329 /* Update stats. (This is a no-op if the rule expired due to an idle
4330 * timeout, because that only happens when the rule has no facets left.) */
4331 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4332 facet_remove(facet);
4335 /* Get rid of the rule. */
4336 ofproto_rule_expire(&rule->up, reason);
4341 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
4343 * The caller must already have determined that no facet with an identical
4344 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
4345 * the ofproto's classifier table.
4347 * 'hash' must be the return value of flow_hash(flow, 0).
4349 * The facet will initially have no subfacets. The caller should create (at
4350 * least) one subfacet with subfacet_create(). */
4351 static struct facet *
4352 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
4354 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4355 struct facet *facet;
4357 facet = xzalloc(sizeof *facet);
4358 facet->used = time_msec();
4359 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4360 list_push_back(&rule->facets, &facet->list_node);
4362 facet->flow = *flow;
4363 list_init(&facet->subfacets);
4364 netflow_flow_init(&facet->nf_flow);
4365 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4371 facet_free(struct facet *facet)
4376 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4377 * 'packet', which arrived on 'in_port'. */
4379 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4380 const struct nlattr *odp_actions, size_t actions_len,
4381 struct ofpbuf *packet)
4383 struct odputil_keybuf keybuf;
4387 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4388 odp_flow_key_from_flow(&key, flow,
4389 ofp_port_to_odp_port(ofproto, flow->in_port));
4391 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4392 odp_actions, actions_len, packet);
4396 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4398 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4399 * rule's statistics, via subfacet_uninstall().
4401 * - Removes 'facet' from its rule and from ofproto->facets.
4404 facet_remove(struct facet *facet)
4406 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4407 struct subfacet *subfacet, *next_subfacet;
4409 ovs_assert(!list_is_empty(&facet->subfacets));
4411 /* First uninstall all of the subfacets to get final statistics. */
4412 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4413 subfacet_uninstall(subfacet);
4416 /* Flush the final stats to the rule.
4418 * This might require us to have at least one subfacet around so that we
4419 * can use its actions for accounting in facet_account(), which is why we
4420 * have uninstalled but not yet destroyed the subfacets. */
4421 facet_flush_stats(facet);
4423 /* Now we're really all done so destroy everything. */
4424 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4425 &facet->subfacets) {
4426 subfacet_destroy__(subfacet);
4428 hmap_remove(&ofproto->facets, &facet->hmap_node);
4429 list_remove(&facet->list_node);
4433 /* Feed information from 'facet' back into the learning table to keep it in
4434 * sync with what is actually flowing through the datapath. */
4436 facet_learn(struct facet *facet)
4438 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4439 struct action_xlate_ctx ctx;
4441 if (!facet->has_learn
4442 && !facet->has_normal
4443 && (!facet->has_fin_timeout
4444 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4448 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4449 facet->flow.vlan_tci,
4450 facet->rule, facet->tcp_flags, NULL);
4451 ctx.may_learn = true;
4452 xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
4453 facet->rule->up.ofpacts_len);
4457 facet_account(struct facet *facet)
4459 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4460 struct subfacet *subfacet;
4461 const struct nlattr *a;
4466 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
4469 n_bytes = facet->byte_count - facet->accounted_bytes;
4471 /* This loop feeds byte counters to bond_account() for rebalancing to use
4472 * as a basis. We also need to track the actual VLAN on which the packet
4473 * is going to be sent to ensure that it matches the one passed to
4474 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4477 * We use the actions from an arbitrary subfacet because they should all
4478 * be equally valid for our purpose. */
4479 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
4480 struct subfacet, list_node);
4481 vlan_tci = facet->flow.vlan_tci;
4482 NL_ATTR_FOR_EACH_UNSAFE (a, left,
4483 subfacet->actions, subfacet->actions_len) {
4484 const struct ovs_action_push_vlan *vlan;
4485 struct ofport_dpif *port;
4487 switch (nl_attr_type(a)) {
4488 case OVS_ACTION_ATTR_OUTPUT:
4489 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4490 if (port && port->bundle && port->bundle->bond) {
4491 bond_account(port->bundle->bond, &facet->flow,
4492 vlan_tci_to_vid(vlan_tci), n_bytes);
4496 case OVS_ACTION_ATTR_POP_VLAN:
4497 vlan_tci = htons(0);
4500 case OVS_ACTION_ATTR_PUSH_VLAN:
4501 vlan = nl_attr_get(a);
4502 vlan_tci = vlan->vlan_tci;
4508 /* Returns true if the only action for 'facet' is to send to the controller.
4509 * (We don't report NetFlow expiration messages for such facets because they
4510 * are just part of the control logic for the network, not real traffic). */
4512 facet_is_controller_flow(struct facet *facet)
4515 const struct rule *rule = &facet->rule->up;
4516 const struct ofpact *ofpacts = rule->ofpacts;
4517 size_t ofpacts_len = rule->ofpacts_len;
4519 if (ofpacts_len > 0 &&
4520 ofpacts->type == OFPACT_CONTROLLER &&
4521 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4528 /* Folds all of 'facet''s statistics into its rule. Also updates the
4529 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4530 * 'facet''s statistics in the datapath should have been zeroed and folded into
4531 * its packet and byte counts before this function is called. */
4533 facet_flush_stats(struct facet *facet)
4535 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4536 struct subfacet *subfacet;
4538 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4539 ovs_assert(!subfacet->dp_byte_count);
4540 ovs_assert(!subfacet->dp_packet_count);
4543 facet_push_stats(facet);
4544 if (facet->accounted_bytes < facet->byte_count) {
4545 facet_account(facet);
4546 facet->accounted_bytes = facet->byte_count;
4549 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4550 struct ofexpired expired;
4551 expired.flow = facet->flow;
4552 expired.packet_count = facet->packet_count;
4553 expired.byte_count = facet->byte_count;
4554 expired.used = facet->used;
4555 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4558 facet->rule->packet_count += facet->packet_count;
4559 facet->rule->byte_count += facet->byte_count;
4561 /* Reset counters to prevent double counting if 'facet' ever gets
4563 facet_reset_counters(facet);
4565 netflow_flow_clear(&facet->nf_flow);
4566 facet->tcp_flags = 0;
4569 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4570 * Returns it if found, otherwise a null pointer.
4572 * 'hash' must be the return value of flow_hash(flow, 0).
4574 * The returned facet might need revalidation; use facet_lookup_valid()
4575 * instead if that is important. */
4576 static struct facet *
4577 facet_find(struct ofproto_dpif *ofproto,
4578 const struct flow *flow, uint32_t hash)
4580 struct facet *facet;
4582 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4583 if (flow_equal(flow, &facet->flow)) {
4591 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4592 * Returns it if found, otherwise a null pointer.
4594 * 'hash' must be the return value of flow_hash(flow, 0).
4596 * The returned facet is guaranteed to be valid. */
4597 static struct facet *
4598 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4601 struct facet *facet;
4603 facet = facet_find(ofproto, flow, hash);
4605 && (ofproto->backer->need_revalidate
4606 || tag_set_intersects(&ofproto->backer->revalidate_set,
4608 facet_revalidate(facet);
4615 subfacet_path_to_string(enum subfacet_path path)
4618 case SF_NOT_INSTALLED:
4619 return "not installed";
4621 return "in fast path";
4623 return "in slow path";
4629 /* Returns the path in which a subfacet should be installed if its 'slow'
4630 * member has the specified value. */
4631 static enum subfacet_path
4632 subfacet_want_path(enum slow_path_reason slow)
4634 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
4637 /* Returns true if 'subfacet' needs to have its datapath flow updated,
4638 * supposing that its actions have been recalculated as 'want_actions' and that
4639 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
4641 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
4642 const struct ofpbuf *want_actions)
4644 enum subfacet_path want_path = subfacet_want_path(slow);
4645 return (want_path != subfacet->path
4646 || (want_path == SF_FAST_PATH
4647 && (subfacet->actions_len != want_actions->size
4648 || memcmp(subfacet->actions, want_actions->data,
4649 subfacet->actions_len))));
4653 facet_check_consistency(struct facet *facet)
4655 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4657 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4659 uint64_t odp_actions_stub[1024 / 8];
4660 struct ofpbuf odp_actions;
4662 struct rule_dpif *rule;
4663 struct subfacet *subfacet;
4664 bool may_log = false;
4667 /* Check the rule for consistency. */
4668 rule = rule_dpif_lookup(ofproto, &facet->flow);
4669 ok = rule == facet->rule;
4671 may_log = !VLOG_DROP_WARN(&rl);
4676 flow_format(&s, &facet->flow);
4677 ds_put_format(&s, ": facet associated with wrong rule (was "
4678 "table=%"PRIu8",", facet->rule->up.table_id);
4679 cls_rule_format(&facet->rule->up.cr, &s);
4680 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4682 cls_rule_format(&rule->up.cr, &s);
4683 ds_put_char(&s, ')');
4685 VLOG_WARN("%s", ds_cstr(&s));
4690 /* Check the datapath actions for consistency. */
4691 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4692 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4693 enum subfacet_path want_path;
4694 struct odputil_keybuf keybuf;
4695 struct action_xlate_ctx ctx;
4699 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4700 subfacet->initial_tci, rule, 0, NULL);
4701 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
4704 if (subfacet->path == SF_NOT_INSTALLED) {
4705 /* This only happens if the datapath reported an error when we
4706 * tried to install the flow. Don't flag another error here. */
4710 want_path = subfacet_want_path(subfacet->slow);
4711 if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) {
4712 /* The actions for slow-path flows may legitimately vary from one
4713 * packet to the next. We're done. */
4717 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
4721 /* Inconsistency! */
4723 may_log = !VLOG_DROP_WARN(&rl);
4727 /* Rate-limited, skip reporting. */
4732 subfacet_get_key(subfacet, &keybuf, &key);
4733 odp_flow_key_format(key.data, key.size, &s);
4735 ds_put_cstr(&s, ": inconsistency in subfacet");
4736 if (want_path != subfacet->path) {
4737 enum odp_key_fitness fitness = subfacet->key_fitness;
4739 ds_put_format(&s, " (%s, fitness=%s)",
4740 subfacet_path_to_string(subfacet->path),
4741 odp_key_fitness_to_string(fitness));
4742 ds_put_format(&s, " (should have been %s)",
4743 subfacet_path_to_string(want_path));
4744 } else if (want_path == SF_FAST_PATH) {
4745 ds_put_cstr(&s, " (actions were: ");
4746 format_odp_actions(&s, subfacet->actions,
4747 subfacet->actions_len);
4748 ds_put_cstr(&s, ") (correct actions: ");
4749 format_odp_actions(&s, odp_actions.data, odp_actions.size);
4750 ds_put_char(&s, ')');
4752 ds_put_cstr(&s, " (actions: ");
4753 format_odp_actions(&s, subfacet->actions,
4754 subfacet->actions_len);
4755 ds_put_char(&s, ')');
4757 VLOG_WARN("%s", ds_cstr(&s));
4760 ofpbuf_uninit(&odp_actions);
4765 /* Re-searches the classifier for 'facet':
4767 * - If the rule found is different from 'facet''s current rule, moves
4768 * 'facet' to the new rule and recompiles its actions.
4770 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4771 * where it is and recompiles its actions anyway. */
4773 facet_revalidate(struct facet *facet)
4775 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4777 struct nlattr *odp_actions;
4780 struct actions *new_actions;
4782 struct action_xlate_ctx ctx;
4783 uint64_t odp_actions_stub[1024 / 8];
4784 struct ofpbuf odp_actions;
4786 struct rule_dpif *new_rule;
4787 struct subfacet *subfacet;
4790 COVERAGE_INC(facet_revalidate);
4792 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
4794 /* Calculate new datapath actions.
4796 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4797 * emit a NetFlow expiration and, if so, we need to have the old state
4798 * around to properly compose it. */
4800 /* If the datapath actions changed or the installability changed,
4801 * then we need to talk to the datapath. */
4804 memset(&ctx, 0, sizeof ctx);
4805 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4806 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4807 enum slow_path_reason slow;
4809 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4810 subfacet->initial_tci, new_rule, 0, NULL);
4811 xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len,
4814 slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4815 if (subfacet_should_install(subfacet, slow, &odp_actions)) {
4816 struct dpif_flow_stats stats;
4818 subfacet_install(subfacet,
4819 odp_actions.data, odp_actions.size, &stats, slow);
4820 subfacet_update_stats(subfacet, &stats);
4823 new_actions = xcalloc(list_size(&facet->subfacets),
4824 sizeof *new_actions);
4826 new_actions[i].odp_actions = xmemdup(odp_actions.data,
4828 new_actions[i].actions_len = odp_actions.size;
4833 ofpbuf_uninit(&odp_actions);
4836 facet_flush_stats(facet);
4839 /* Update 'facet' now that we've taken care of all the old state. */
4840 facet->tags = ctx.tags;
4841 facet->nf_flow.output_iface = ctx.nf_output_iface;
4842 facet->has_learn = ctx.has_learn;
4843 facet->has_normal = ctx.has_normal;
4844 facet->has_fin_timeout = ctx.has_fin_timeout;
4845 facet->mirrors = ctx.mirrors;
4848 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4849 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4851 if (new_actions && new_actions[i].odp_actions) {
4852 free(subfacet->actions);
4853 subfacet->actions = new_actions[i].odp_actions;
4854 subfacet->actions_len = new_actions[i].actions_len;
4860 if (facet->rule != new_rule) {
4861 COVERAGE_INC(facet_changed_rule);
4862 list_remove(&facet->list_node);
4863 list_push_back(&new_rule->facets, &facet->list_node);
4864 facet->rule = new_rule;
4865 facet->used = new_rule->up.created;
4866 facet->prev_used = facet->used;
4870 /* Updates 'facet''s used time. Caller is responsible for calling
4871 * facet_push_stats() to update the flows which 'facet' resubmits into. */
4873 facet_update_time(struct facet *facet, long long int used)
4875 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4876 if (used > facet->used) {
4878 ofproto_rule_update_used(&facet->rule->up, used);
4879 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
4884 facet_reset_counters(struct facet *facet)
4886 facet->packet_count = 0;
4887 facet->byte_count = 0;
4888 facet->prev_packet_count = 0;
4889 facet->prev_byte_count = 0;
4890 facet->accounted_bytes = 0;
4894 facet_push_stats(struct facet *facet)
4896 struct dpif_flow_stats stats;
4898 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4899 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4900 ovs_assert(facet->used >= facet->prev_used);
4902 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4903 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4904 stats.used = facet->used;
4905 stats.tcp_flags = 0;
4907 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
4908 facet->prev_packet_count = facet->packet_count;
4909 facet->prev_byte_count = facet->byte_count;
4910 facet->prev_used = facet->used;
4912 flow_push_stats(facet->rule, &facet->flow, &stats);
4914 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
4915 facet->mirrors, stats.n_packets, stats.n_bytes);
4920 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4922 rule->packet_count += stats->n_packets;
4923 rule->byte_count += stats->n_bytes;
4924 ofproto_rule_update_used(&rule->up, stats->used);
4927 /* Pushes flow statistics to the rules which 'flow' resubmits into given
4928 * 'rule''s actions and mirrors. */
4930 flow_push_stats(struct rule_dpif *rule,
4931 const struct flow *flow, const struct dpif_flow_stats *stats)
4933 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4934 struct action_xlate_ctx ctx;
4936 ofproto_rule_update_used(&rule->up, stats->used);
4938 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, rule,
4940 ctx.resubmit_stats = stats;
4941 xlate_actions_for_side_effects(&ctx, rule->up.ofpacts,
4942 rule->up.ofpacts_len);
4947 static struct subfacet *
4948 subfacet_find(struct ofproto_dpif *ofproto,
4949 const struct nlattr *key, size_t key_len, uint32_t key_hash,
4950 const struct flow *flow)
4952 struct subfacet *subfacet;
4954 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4955 &ofproto->subfacets) {
4957 ? (subfacet->key_len == key_len
4958 && !memcmp(key, subfacet->key, key_len))
4959 : flow_equal(flow, &subfacet->facet->flow)) {
4967 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4968 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4969 * existing subfacet if there is one, otherwise creates and returns a
4972 * If the returned subfacet is new, then subfacet->actions will be NULL, in
4973 * which case the caller must populate the actions with
4974 * subfacet_make_actions(). */
4975 static struct subfacet *
4976 subfacet_create(struct facet *facet, struct flow_miss *miss,
4979 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4980 enum odp_key_fitness key_fitness = miss->key_fitness;
4981 const struct nlattr *key = miss->key;
4982 size_t key_len = miss->key_len;
4984 struct subfacet *subfacet;
4986 key_hash = odp_flow_key_hash(key, key_len);
4988 if (list_is_empty(&facet->subfacets)) {
4989 subfacet = &facet->one_subfacet;
4991 subfacet = subfacet_find(ofproto, key, key_len, key_hash,
4994 if (subfacet->facet == facet) {
4998 /* This shouldn't happen. */
4999 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
5000 subfacet_destroy(subfacet);
5003 subfacet = xmalloc(sizeof *subfacet);
5006 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
5007 list_push_back(&facet->subfacets, &subfacet->list_node);
5008 subfacet->facet = facet;
5009 subfacet->key_fitness = key_fitness;
5010 if (key_fitness != ODP_FIT_PERFECT) {
5011 subfacet->key = xmemdup(key, key_len);
5012 subfacet->key_len = key_len;
5014 subfacet->key = NULL;
5015 subfacet->key_len = 0;
5017 subfacet->used = now;
5018 subfacet->dp_packet_count = 0;
5019 subfacet->dp_byte_count = 0;
5020 subfacet->actions_len = 0;
5021 subfacet->actions = NULL;
5022 subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE
5025 subfacet->path = SF_NOT_INSTALLED;
5026 subfacet->initial_tci = miss->initial_tci;
5027 subfacet->odp_in_port = miss->odp_in_port;
5032 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
5033 * its facet within 'ofproto', and frees it. */
5035 subfacet_destroy__(struct subfacet *subfacet)
5037 struct facet *facet = subfacet->facet;
5038 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5040 subfacet_uninstall(subfacet);
5041 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
5042 list_remove(&subfacet->list_node);
5043 free(subfacet->key);
5044 free(subfacet->actions);
5045 if (subfacet != &facet->one_subfacet) {
5050 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5051 * last remaining subfacet in its facet destroys the facet too. */
5053 subfacet_destroy(struct subfacet *subfacet)
5055 struct facet *facet = subfacet->facet;
5057 if (list_is_singleton(&facet->subfacets)) {
5058 /* facet_remove() needs at least one subfacet (it will remove it). */
5059 facet_remove(facet);
5061 subfacet_destroy__(subfacet);
5066 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
5067 struct subfacet **subfacets, int n)
5069 struct odputil_keybuf keybufs[SUBFACET_DESTROY_MAX_BATCH];
5070 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5071 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5072 struct ofpbuf keys[SUBFACET_DESTROY_MAX_BATCH];
5073 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5076 for (i = 0; i < n; i++) {
5077 ops[i].type = DPIF_OP_FLOW_DEL;
5078 subfacet_get_key(subfacets[i], &keybufs[i], &keys[i]);
5079 ops[i].u.flow_del.key = keys[i].data;
5080 ops[i].u.flow_del.key_len = keys[i].size;
5081 ops[i].u.flow_del.stats = &stats[i];
5085 dpif_operate(ofproto->backer->dpif, opsp, n);
5086 for (i = 0; i < n; i++) {
5087 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5088 subfacets[i]->path = SF_NOT_INSTALLED;
5089 subfacet_destroy(subfacets[i]);
5093 /* Initializes 'key' with the sequence of OVS_KEY_ATTR_* Netlink attributes
5094 * that can be used to refer to 'subfacet'. The caller must provide 'keybuf'
5095 * for use as temporary storage. */
5097 subfacet_get_key(struct subfacet *subfacet, struct odputil_keybuf *keybuf,
5101 if (!subfacet->key) {
5102 struct flow *flow = &subfacet->facet->flow;
5104 ofpbuf_use_stack(key, keybuf, sizeof *keybuf);
5105 odp_flow_key_from_flow(key, flow, subfacet->odp_in_port);
5107 ofpbuf_use_const(key, subfacet->key, subfacet->key_len);
5111 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
5112 * Translates the actions into 'odp_actions', which the caller must have
5113 * initialized and is responsible for uninitializing. */
5115 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
5116 struct ofpbuf *odp_actions)
5118 struct facet *facet = subfacet->facet;
5119 struct rule_dpif *rule = facet->rule;
5120 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5122 struct action_xlate_ctx ctx;
5124 action_xlate_ctx_init(&ctx, ofproto, &facet->flow, subfacet->initial_tci,
5126 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions);
5127 facet->tags = ctx.tags;
5128 facet->has_learn = ctx.has_learn;
5129 facet->has_normal = ctx.has_normal;
5130 facet->has_fin_timeout = ctx.has_fin_timeout;
5131 facet->nf_flow.output_iface = ctx.nf_output_iface;
5132 facet->mirrors = ctx.mirrors;
5134 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
5135 if (subfacet->actions_len != odp_actions->size
5136 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
5137 free(subfacet->actions);
5138 subfacet->actions_len = odp_actions->size;
5139 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
5143 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5144 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5145 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5146 * since 'subfacet' was last updated.
5148 * Returns 0 if successful, otherwise a positive errno value. */
5150 subfacet_install(struct subfacet *subfacet,
5151 const struct nlattr *actions, size_t actions_len,
5152 struct dpif_flow_stats *stats,
5153 enum slow_path_reason slow)
5155 struct facet *facet = subfacet->facet;
5156 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5157 enum subfacet_path path = subfacet_want_path(slow);
5158 uint64_t slow_path_stub[128 / 8];
5159 struct odputil_keybuf keybuf;
5160 enum dpif_flow_put_flags flags;
5164 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5166 flags |= DPIF_FP_ZERO_STATS;
5169 if (path == SF_SLOW_PATH) {
5170 compose_slow_path(ofproto, &facet->flow, slow,
5171 slow_path_stub, sizeof slow_path_stub,
5172 &actions, &actions_len);
5175 subfacet_get_key(subfacet, &keybuf, &key);
5176 ret = dpif_flow_put(ofproto->backer->dpif, flags, key.data, key.size,
5177 actions, actions_len, stats);
5180 subfacet_reset_dp_stats(subfacet, stats);
5184 subfacet->path = path;
5190 subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats)
5192 return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len,
5193 stats, subfacet->slow);
5196 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5198 subfacet_uninstall(struct subfacet *subfacet)
5200 if (subfacet->path != SF_NOT_INSTALLED) {
5201 struct rule_dpif *rule = subfacet->facet->rule;
5202 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5203 struct odputil_keybuf keybuf;
5204 struct dpif_flow_stats stats;
5208 subfacet_get_key(subfacet, &keybuf, &key);
5209 error = dpif_flow_del(ofproto->backer->dpif,
5210 key.data, key.size, &stats);
5211 subfacet_reset_dp_stats(subfacet, &stats);
5213 subfacet_update_stats(subfacet, &stats);
5215 subfacet->path = SF_NOT_INSTALLED;
5217 ovs_assert(subfacet->dp_packet_count == 0);
5218 ovs_assert(subfacet->dp_byte_count == 0);
5222 /* Resets 'subfacet''s datapath statistics counters. This should be called
5223 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5224 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5225 * was reset in the datapath. 'stats' will be modified to include only
5226 * statistics new since 'subfacet' was last updated. */
5228 subfacet_reset_dp_stats(struct subfacet *subfacet,
5229 struct dpif_flow_stats *stats)
5232 && subfacet->dp_packet_count <= stats->n_packets
5233 && subfacet->dp_byte_count <= stats->n_bytes) {
5234 stats->n_packets -= subfacet->dp_packet_count;
5235 stats->n_bytes -= subfacet->dp_byte_count;
5238 subfacet->dp_packet_count = 0;
5239 subfacet->dp_byte_count = 0;
5242 /* Updates 'subfacet''s used time. The caller is responsible for calling
5243 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
5245 subfacet_update_time(struct subfacet *subfacet, long long int used)
5247 if (used > subfacet->used) {
5248 subfacet->used = used;
5249 facet_update_time(subfacet->facet, used);
5253 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5255 * Because of the meaning of a subfacet's counters, it only makes sense to do
5256 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5257 * represents a packet that was sent by hand or if it represents statistics
5258 * that have been cleared out of the datapath. */
5260 subfacet_update_stats(struct subfacet *subfacet,
5261 const struct dpif_flow_stats *stats)
5263 if (stats->n_packets || stats->used > subfacet->used) {
5264 struct facet *facet = subfacet->facet;
5266 subfacet_update_time(subfacet, stats->used);
5267 facet->packet_count += stats->n_packets;
5268 facet->byte_count += stats->n_bytes;
5269 facet->tcp_flags |= stats->tcp_flags;
5270 facet_push_stats(facet);
5271 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
5277 static struct rule_dpif *
5278 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
5280 struct rule_dpif *rule;
5282 rule = rule_dpif_lookup__(ofproto, flow, 0);
5287 return rule_dpif_miss_rule(ofproto, flow);
5290 static struct rule_dpif *
5291 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5294 struct cls_rule *cls_rule;
5295 struct classifier *cls;
5297 if (table_id >= N_TABLES) {
5301 cls = &ofproto->up.tables[table_id].cls;
5302 if (flow->nw_frag & FLOW_NW_FRAG_ANY
5303 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5304 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
5305 * are unavailable. */
5306 struct flow ofpc_normal_flow = *flow;
5307 ofpc_normal_flow.tp_src = htons(0);
5308 ofpc_normal_flow.tp_dst = htons(0);
5309 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
5311 cls_rule = classifier_lookup(cls, flow);
5313 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5316 static struct rule_dpif *
5317 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5319 struct ofport_dpif *port;
5321 port = get_ofp_port(ofproto, flow->in_port);
5323 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5324 return ofproto->miss_rule;
5327 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5328 return ofproto->no_packet_in_rule;
5330 return ofproto->miss_rule;
5334 complete_operation(struct rule_dpif *rule)
5336 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5338 rule_invalidate(rule);
5340 struct dpif_completion *c = xmalloc(sizeof *c);
5341 c->op = rule->up.pending;
5342 list_push_back(&ofproto->completions, &c->list_node);
5344 ofoperation_complete(rule->up.pending, 0);
5348 static struct rule *
5351 struct rule_dpif *rule = xmalloc(sizeof *rule);
5356 rule_dealloc(struct rule *rule_)
5358 struct rule_dpif *rule = rule_dpif_cast(rule_);
5363 rule_construct(struct rule *rule_)
5365 struct rule_dpif *rule = rule_dpif_cast(rule_);
5366 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5367 struct rule_dpif *victim;
5370 rule->packet_count = 0;
5371 rule->byte_count = 0;
5373 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5374 if (victim && !list_is_empty(&victim->facets)) {
5375 struct facet *facet;
5377 rule->facets = victim->facets;
5378 list_moved(&rule->facets);
5379 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5380 /* XXX: We're only clearing our local counters here. It's possible
5381 * that quite a few packets are unaccounted for in the datapath
5382 * statistics. These will be accounted to the new rule instead of
5383 * cleared as required. This could be fixed by clearing out the
5384 * datapath statistics for this facet, but currently it doesn't
5386 facet_reset_counters(facet);
5390 /* Must avoid list_moved() in this case. */
5391 list_init(&rule->facets);
5394 table_id = rule->up.table_id;
5396 rule->tag = victim->tag;
5397 } else if (table_id == 0) {
5402 miniflow_expand(&rule->up.cr.match.flow, &flow);
5403 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5404 ofproto->tables[table_id].basis);
5407 complete_operation(rule);
5412 rule_destruct(struct rule *rule_)
5414 struct rule_dpif *rule = rule_dpif_cast(rule_);
5415 struct facet *facet, *next_facet;
5417 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5418 facet_revalidate(facet);
5421 complete_operation(rule);
5425 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5427 struct rule_dpif *rule = rule_dpif_cast(rule_);
5428 struct facet *facet;
5430 /* Start from historical data for 'rule' itself that are no longer tracked
5431 * in facets. This counts, for example, facets that have expired. */
5432 *packets = rule->packet_count;
5433 *bytes = rule->byte_count;
5435 /* Add any statistics that are tracked by facets. This includes
5436 * statistical data recently updated by ofproto_update_stats() as well as
5437 * stats for packets that were executed "by hand" via dpif_execute(). */
5438 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5439 *packets += facet->packet_count;
5440 *bytes += facet->byte_count;
5445 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5446 struct ofpbuf *packet)
5448 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5450 struct dpif_flow_stats stats;
5452 struct action_xlate_ctx ctx;
5453 uint64_t odp_actions_stub[1024 / 8];
5454 struct ofpbuf odp_actions;
5456 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5457 rule_credit_stats(rule, &stats);
5459 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5460 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci,
5461 rule, stats.tcp_flags, packet);
5462 ctx.resubmit_stats = &stats;
5463 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
5465 execute_odp_actions(ofproto, flow, odp_actions.data,
5466 odp_actions.size, packet);
5468 ofpbuf_uninit(&odp_actions);
5472 rule_execute(struct rule *rule, const struct flow *flow,
5473 struct ofpbuf *packet)
5475 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5476 ofpbuf_delete(packet);
5481 rule_modify_actions(struct rule *rule_)
5483 struct rule_dpif *rule = rule_dpif_cast(rule_);
5485 complete_operation(rule);
5488 /* Sends 'packet' out 'ofport'.
5489 * May modify 'packet'.
5490 * Returns 0 if successful, otherwise a positive errno value. */
5492 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5494 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5495 uint64_t odp_actions_stub[1024 / 8];
5496 struct ofpbuf key, odp_actions;
5497 struct odputil_keybuf keybuf;
5502 flow_extract(packet, 0, 0, NULL, OFPP_LOCAL, &flow);
5503 if (netdev_vport_is_patch(ofport->up.netdev)) {
5504 struct ofproto_dpif *peer_ofproto;
5505 struct dpif_flow_stats stats;
5506 struct ofport_dpif *peer;
5507 struct rule_dpif *rule;
5509 peer = ofport_get_peer(ofport);
5514 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5515 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5516 netdev_vport_inc_rx(peer->up.netdev, &stats);
5518 flow.in_port = peer->up.ofp_port;
5519 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5520 rule = rule_dpif_lookup(peer_ofproto, &flow);
5521 rule_dpif_execute(rule, &flow, packet);
5526 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5528 if (ofport->tnl_port) {
5529 struct dpif_flow_stats stats;
5531 odp_port = tnl_port_send(ofport->tnl_port, &flow);
5532 if (odp_port == OVSP_NONE) {
5536 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5537 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5538 odp_put_tunnel_action(&flow.tunnel, &odp_actions);
5539 odp_put_skb_mark_action(flow.skb_mark, &odp_actions);
5541 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
5543 if (odp_port != ofport->odp_port) {
5544 eth_pop_vlan(packet);
5545 flow.vlan_tci = htons(0);
5549 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5550 odp_flow_key_from_flow(&key, &flow,
5551 ofp_port_to_odp_port(ofproto, flow.in_port));
5553 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
5555 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
5556 error = dpif_execute(ofproto->backer->dpif,
5558 odp_actions.data, odp_actions.size,
5560 ofpbuf_uninit(&odp_actions);
5563 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
5564 ofproto->up.name, odp_port, strerror(error));
5566 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
5570 /* OpenFlow to datapath action translation. */
5572 static bool may_receive(const struct ofport_dpif *, struct action_xlate_ctx *);
5573 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5574 struct action_xlate_ctx *);
5575 static void xlate_normal(struct action_xlate_ctx *);
5577 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5578 * The action will state 'slow' as the reason that the action is in the slow
5579 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5580 * dump-flows" output to see why a flow is in the slow path.)
5582 * The 'stub_size' bytes in 'stub' will be used to store the action.
5583 * 'stub_size' must be large enough for the action.
5585 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5588 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5589 enum slow_path_reason slow,
5590 uint64_t *stub, size_t stub_size,
5591 const struct nlattr **actionsp, size_t *actions_lenp)
5593 union user_action_cookie cookie;
5596 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5597 cookie.slow_path.unused = 0;
5598 cookie.slow_path.reason = slow;
5600 ofpbuf_use_stack(&buf, stub, stub_size);
5601 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
5602 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5603 odp_put_userspace_action(pid, &cookie, &buf);
5605 put_userspace_action(ofproto, &buf, flow, &cookie);
5607 *actionsp = buf.data;
5608 *actions_lenp = buf.size;
5612 put_userspace_action(const struct ofproto_dpif *ofproto,
5613 struct ofpbuf *odp_actions,
5614 const struct flow *flow,
5615 const union user_action_cookie *cookie)
5619 pid = dpif_port_get_pid(ofproto->backer->dpif,
5620 ofp_port_to_odp_port(ofproto, flow->in_port));
5622 return odp_put_userspace_action(pid, cookie, odp_actions);
5626 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5627 ovs_be16 vlan_tci, uint32_t odp_port,
5628 unsigned int n_outputs, union user_action_cookie *cookie)
5632 cookie->type = USER_ACTION_COOKIE_SFLOW;
5633 cookie->sflow.vlan_tci = vlan_tci;
5635 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5636 * port information") for the interpretation of cookie->output. */
5637 switch (n_outputs) {
5639 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5640 cookie->sflow.output = 0x40000000 | 256;
5644 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5646 cookie->sflow.output = ifindex;
5651 /* 0x80000000 means "multiple output ports. */
5652 cookie->sflow.output = 0x80000000 | n_outputs;
5657 /* Compose SAMPLE action for sFlow. */
5659 compose_sflow_action(const struct ofproto_dpif *ofproto,
5660 struct ofpbuf *odp_actions,
5661 const struct flow *flow,
5664 uint32_t probability;
5665 union user_action_cookie cookie;
5666 size_t sample_offset, actions_offset;
5669 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5673 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5675 /* Number of packets out of UINT_MAX to sample. */
5676 probability = dpif_sflow_get_probability(ofproto->sflow);
5677 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5679 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5680 compose_sflow_cookie(ofproto, htons(0), odp_port,
5681 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5682 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
5684 nl_msg_end_nested(odp_actions, actions_offset);
5685 nl_msg_end_nested(odp_actions, sample_offset);
5686 return cookie_offset;
5689 /* SAMPLE action must be first action in any given list of actions.
5690 * At this point we do not have all information required to build it. So try to
5691 * build sample action as complete as possible. */
5693 add_sflow_action(struct action_xlate_ctx *ctx)
5695 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
5697 &ctx->flow, OVSP_NONE);
5698 ctx->sflow_odp_port = 0;
5699 ctx->sflow_n_outputs = 0;
5702 /* Fix SAMPLE action according to data collected while composing ODP actions.
5703 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
5704 * USERSPACE action's user-cookie which is required for sflow. */
5706 fix_sflow_action(struct action_xlate_ctx *ctx)
5708 const struct flow *base = &ctx->base_flow;
5709 union user_action_cookie *cookie;
5711 if (!ctx->user_cookie_offset) {
5715 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
5717 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
5719 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
5720 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
5724 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
5727 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
5728 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
5729 ovs_be64 flow_tun_id = ctx->flow.tunnel.tun_id;
5730 uint8_t flow_nw_tos = ctx->flow.nw_tos;
5731 struct priority_to_dscp *pdscp;
5732 uint32_t out_port, odp_port;
5734 /* If 'struct flow' gets additional metadata, we'll need to zero it out
5735 * before traversing a patch port. */
5736 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 18);
5739 xlate_report(ctx, "Nonexistent output port");
5741 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
5742 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
5744 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
5745 xlate_report(ctx, "STP not in forwarding state, skipping output");
5749 if (netdev_vport_is_patch(ofport->up.netdev)) {
5750 struct ofport_dpif *peer = ofport_get_peer(ofport);
5751 struct flow old_flow = ctx->flow;
5752 const struct ofproto_dpif *peer_ofproto;
5753 enum slow_path_reason special;
5754 struct ofport_dpif *in_port;
5757 xlate_report(ctx, "Nonexistent patch port peer");
5761 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5762 if (peer_ofproto->backer != ctx->ofproto->backer) {
5763 xlate_report(ctx, "Patch port peer on a different datapath");
5767 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
5768 ctx->flow.in_port = peer->up.ofp_port;
5769 ctx->flow.metadata = htonll(0);
5770 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
5771 memset(ctx->flow.regs, 0, sizeof ctx->flow.regs);
5773 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5774 special = process_special(ctx->ofproto, &ctx->flow, in_port,
5777 ctx->slow |= special;
5778 } else if (!in_port || may_receive(in_port, ctx)) {
5779 if (!in_port || stp_forward_in_state(in_port->stp_state)) {
5780 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
5782 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
5783 * learning action look at the packet, then drop it. */
5784 struct flow old_base_flow = ctx->base_flow;
5785 size_t old_size = ctx->odp_actions->size;
5786 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
5787 ctx->base_flow = old_base_flow;
5788 ctx->odp_actions->size = old_size;
5792 ctx->flow = old_flow;
5793 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5795 if (ctx->resubmit_stats) {
5796 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
5797 netdev_vport_inc_rx(peer->up.netdev, ctx->resubmit_stats);
5803 pdscp = get_priority(ofport, ctx->flow.skb_priority);
5805 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5806 ctx->flow.nw_tos |= pdscp->dscp;
5809 odp_port = ofp_port_to_odp_port(ctx->ofproto, ofp_port);
5810 if (ofport->tnl_port) {
5811 odp_port = tnl_port_send(ofport->tnl_port, &ctx->flow);
5812 if (odp_port == OVSP_NONE) {
5813 xlate_report(ctx, "Tunneling decided against output");
5817 if (ctx->resubmit_stats) {
5818 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
5820 out_port = odp_port;
5821 commit_odp_tunnel_action(&ctx->flow, &ctx->base_flow,
5824 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
5825 ctx->flow.vlan_tci);
5826 if (out_port != odp_port) {
5827 ctx->flow.vlan_tci = htons(0);
5830 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
5831 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
5833 ctx->sflow_odp_port = odp_port;
5834 ctx->sflow_n_outputs++;
5835 ctx->nf_output_iface = ofp_port;
5836 ctx->flow.tunnel.tun_id = flow_tun_id;
5837 ctx->flow.vlan_tci = flow_vlan_tci;
5838 ctx->flow.nw_tos = flow_nw_tos;
5842 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
5844 compose_output_action__(ctx, ofp_port, true);
5848 xlate_table_action(struct action_xlate_ctx *ctx,
5849 uint16_t in_port, uint8_t table_id, bool may_packet_in)
5851 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
5852 struct ofproto_dpif *ofproto = ctx->ofproto;
5853 struct rule_dpif *rule;
5854 uint16_t old_in_port;
5855 uint8_t old_table_id;
5857 old_table_id = ctx->table_id;
5858 ctx->table_id = table_id;
5860 /* Look up a flow with 'in_port' as the input port. */
5861 old_in_port = ctx->flow.in_port;
5862 ctx->flow.in_port = in_port;
5863 rule = rule_dpif_lookup__(ofproto, &ctx->flow, table_id);
5866 if (table_id > 0 && table_id < N_TABLES) {
5867 struct table_dpif *table = &ofproto->tables[table_id];
5868 if (table->other_table) {
5869 ctx->tags |= (rule && rule->tag
5871 : rule_calculate_tag(&ctx->flow,
5872 &table->other_table->mask,
5877 /* Restore the original input port. Otherwise OFPP_NORMAL and
5878 * OFPP_IN_PORT will have surprising behavior. */
5879 ctx->flow.in_port = old_in_port;
5881 if (ctx->resubmit_hook) {
5882 ctx->resubmit_hook(ctx, rule);
5885 if (rule == NULL && may_packet_in) {
5887 * check if table configuration flags
5888 * OFPTC_TABLE_MISS_CONTROLLER, default.
5889 * OFPTC_TABLE_MISS_CONTINUE,
5890 * OFPTC_TABLE_MISS_DROP
5891 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
5893 rule = rule_dpif_miss_rule(ofproto, &ctx->flow);
5897 struct rule_dpif *old_rule = ctx->rule;
5899 if (ctx->resubmit_stats) {
5900 rule_credit_stats(rule, ctx->resubmit_stats);
5905 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
5906 ctx->rule = old_rule;
5910 ctx->table_id = old_table_id;
5912 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5914 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
5915 MAX_RESUBMIT_RECURSION);
5916 ctx->max_resubmit_trigger = true;
5921 xlate_ofpact_resubmit(struct action_xlate_ctx *ctx,
5922 const struct ofpact_resubmit *resubmit)
5927 in_port = resubmit->in_port;
5928 if (in_port == OFPP_IN_PORT) {
5929 in_port = ctx->flow.in_port;
5932 table_id = resubmit->table_id;
5933 if (table_id == 255) {
5934 table_id = ctx->table_id;
5937 xlate_table_action(ctx, in_port, table_id, false);
5941 flood_packets(struct action_xlate_ctx *ctx, bool all)
5943 struct ofport_dpif *ofport;
5945 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
5946 uint16_t ofp_port = ofport->up.ofp_port;
5948 if (ofp_port == ctx->flow.in_port) {
5953 compose_output_action__(ctx, ofp_port, false);
5954 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
5955 compose_output_action(ctx, ofp_port);
5959 ctx->nf_output_iface = NF_OUT_FLOOD;
5963 execute_controller_action(struct action_xlate_ctx *ctx, int len,
5964 enum ofp_packet_in_reason reason,
5965 uint16_t controller_id)
5967 struct ofputil_packet_in pin;
5968 struct ofpbuf *packet;
5970 ctx->slow |= SLOW_CONTROLLER;
5975 packet = ofpbuf_clone(ctx->packet);
5977 if (packet->l2 && packet->l3) {
5978 struct eth_header *eh;
5980 eth_pop_vlan(packet);
5983 /* If the Ethernet type is less than ETH_TYPE_MIN, it's likely an 802.2
5984 * LLC frame. Calculating the Ethernet type of these frames is more
5985 * trouble than seems appropriate for a simple assertion. */
5986 ovs_assert(ntohs(eh->eth_type) < ETH_TYPE_MIN
5987 || eh->eth_type == ctx->flow.dl_type);
5989 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
5990 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
5992 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
5993 eth_push_vlan(packet, ctx->flow.vlan_tci);
5997 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5998 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
5999 ctx->flow.nw_tos, ctx->flow.nw_ttl);
6003 if (ctx->flow.nw_proto == IPPROTO_TCP) {
6004 packet_set_tcp_port(packet, ctx->flow.tp_src,
6006 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
6007 packet_set_udp_port(packet, ctx->flow.tp_src,
6014 pin.packet = packet->data;
6015 pin.packet_len = packet->size;
6016 pin.reason = reason;
6017 pin.controller_id = controller_id;
6018 pin.table_id = ctx->table_id;
6019 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
6022 flow_get_metadata(&ctx->flow, &pin.fmd);
6024 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
6025 ofpbuf_delete(packet);
6029 compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
6031 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
6032 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
6036 if (ctx->flow.nw_ttl > 1) {
6042 for (i = 0; i < ids->n_controllers; i++) {
6043 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
6047 /* Stop processing for current table. */
6053 xlate_output_action(struct action_xlate_ctx *ctx,
6054 uint16_t port, uint16_t max_len, bool may_packet_in)
6056 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
6058 ctx->nf_output_iface = NF_OUT_DROP;
6062 compose_output_action(ctx, ctx->flow.in_port);
6065 xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in);
6071 flood_packets(ctx, false);
6074 flood_packets(ctx, true);
6076 case OFPP_CONTROLLER:
6077 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
6083 if (port != ctx->flow.in_port) {
6084 compose_output_action(ctx, port);
6086 xlate_report(ctx, "skipping output to input port");
6091 if (prev_nf_output_iface == NF_OUT_FLOOD) {
6092 ctx->nf_output_iface = NF_OUT_FLOOD;
6093 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
6094 ctx->nf_output_iface = prev_nf_output_iface;
6095 } else if (prev_nf_output_iface != NF_OUT_DROP &&
6096 ctx->nf_output_iface != NF_OUT_FLOOD) {
6097 ctx->nf_output_iface = NF_OUT_MULTI;
6102 xlate_output_reg_action(struct action_xlate_ctx *ctx,
6103 const struct ofpact_output_reg *or)
6105 uint64_t port = mf_get_subfield(&or->src, &ctx->flow);
6106 if (port <= UINT16_MAX) {
6107 xlate_output_action(ctx, port, or->max_len, false);
6112 xlate_enqueue_action(struct action_xlate_ctx *ctx,
6113 const struct ofpact_enqueue *enqueue)
6115 uint16_t ofp_port = enqueue->port;
6116 uint32_t queue_id = enqueue->queue;
6117 uint32_t flow_priority, priority;
6120 /* Translate queue to priority. */
6121 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6122 queue_id, &priority);
6124 /* Fall back to ordinary output action. */
6125 xlate_output_action(ctx, enqueue->port, 0, false);
6129 /* Check output port. */
6130 if (ofp_port == OFPP_IN_PORT) {
6131 ofp_port = ctx->flow.in_port;
6132 } else if (ofp_port == ctx->flow.in_port) {
6136 /* Add datapath actions. */
6137 flow_priority = ctx->flow.skb_priority;
6138 ctx->flow.skb_priority = priority;
6139 compose_output_action(ctx, ofp_port);
6140 ctx->flow.skb_priority = flow_priority;
6142 /* Update NetFlow output port. */
6143 if (ctx->nf_output_iface == NF_OUT_DROP) {
6144 ctx->nf_output_iface = ofp_port;
6145 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
6146 ctx->nf_output_iface = NF_OUT_MULTI;
6151 xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id)
6153 uint32_t skb_priority;
6155 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6156 queue_id, &skb_priority)) {
6157 ctx->flow.skb_priority = skb_priority;
6159 /* Couldn't translate queue to a priority. Nothing to do. A warning
6160 * has already been logged. */
6164 struct xlate_reg_state {
6170 xlate_autopath(struct action_xlate_ctx *ctx,
6171 const struct ofpact_autopath *ap)
6173 uint16_t ofp_port = ap->port;
6174 struct ofport_dpif *port = get_ofp_port(ctx->ofproto, ofp_port);
6176 if (!port || !port->bundle) {
6177 ofp_port = OFPP_NONE;
6178 } else if (port->bundle->bond) {
6179 /* Autopath does not support VLAN hashing. */
6180 struct ofport_dpif *slave = bond_choose_output_slave(
6181 port->bundle->bond, &ctx->flow, 0, &ctx->tags);
6183 ofp_port = slave->up.ofp_port;
6186 nxm_reg_load(&ap->dst, ofp_port, &ctx->flow);
6190 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
6192 struct ofproto_dpif *ofproto = ofproto_;
6193 struct ofport_dpif *port;
6203 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
6206 port = get_ofp_port(ofproto, ofp_port);
6207 return port ? port->may_enable : false;
6212 xlate_bundle_action(struct action_xlate_ctx *ctx,
6213 const struct ofpact_bundle *bundle)
6217 port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto);
6218 if (bundle->dst.field) {
6219 nxm_reg_load(&bundle->dst, port, &ctx->flow);
6221 xlate_output_action(ctx, port, 0, false);
6226 xlate_learn_action(struct action_xlate_ctx *ctx,
6227 const struct ofpact_learn *learn)
6229 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
6230 struct ofputil_flow_mod fm;
6231 uint64_t ofpacts_stub[1024 / 8];
6232 struct ofpbuf ofpacts;
6235 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
6236 learn_execute(learn, &ctx->flow, &fm, &ofpacts);
6238 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
6239 if (error && !VLOG_DROP_WARN(&rl)) {
6240 VLOG_WARN("learning action failed to modify flow table (%s)",
6241 ofperr_get_name(error));
6244 ofpbuf_uninit(&ofpacts);
6247 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
6248 * means "infinite". */
6250 reduce_timeout(uint16_t max, uint16_t *timeout)
6252 if (max && (!*timeout || *timeout > max)) {
6258 xlate_fin_timeout(struct action_xlate_ctx *ctx,
6259 const struct ofpact_fin_timeout *oft)
6261 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
6262 struct rule_dpif *rule = ctx->rule;
6264 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
6265 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
6270 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
6272 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
6273 ? OFPUTIL_PC_NO_RECV_STP
6274 : OFPUTIL_PC_NO_RECV)) {
6278 /* Only drop packets here if both forwarding and learning are
6279 * disabled. If just learning is enabled, we need to have
6280 * OFPP_NORMAL and the learning action have a look at the packet
6281 * before we can drop it. */
6282 if (!stp_forward_in_state(port->stp_state)
6283 && !stp_learn_in_state(port->stp_state)) {
6291 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
6292 struct action_xlate_ctx *ctx)
6294 bool was_evictable = true;
6295 const struct ofpact *a;
6298 /* Don't let the rule we're working on get evicted underneath us. */
6299 was_evictable = ctx->rule->up.evictable;
6300 ctx->rule->up.evictable = false;
6302 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
6303 struct ofpact_controller *controller;
6304 const struct ofpact_metadata *metadata;
6312 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
6313 ofpact_get_OUTPUT(a)->max_len, true);
6316 case OFPACT_CONTROLLER:
6317 controller = ofpact_get_CONTROLLER(a);
6318 execute_controller_action(ctx, controller->max_len,
6320 controller->controller_id);
6323 case OFPACT_ENQUEUE:
6324 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
6327 case OFPACT_SET_VLAN_VID:
6328 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
6329 ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
6333 case OFPACT_SET_VLAN_PCP:
6334 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
6335 ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
6340 case OFPACT_STRIP_VLAN:
6341 ctx->flow.vlan_tci = htons(0);
6344 case OFPACT_PUSH_VLAN:
6345 /* XXX 802.1AD(QinQ) */
6346 ctx->flow.vlan_tci = htons(VLAN_CFI);
6349 case OFPACT_SET_ETH_SRC:
6350 memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
6354 case OFPACT_SET_ETH_DST:
6355 memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
6359 case OFPACT_SET_IPV4_SRC:
6360 ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
6363 case OFPACT_SET_IPV4_DST:
6364 ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
6367 case OFPACT_SET_IPV4_DSCP:
6368 /* OpenFlow 1.0 only supports IPv4. */
6369 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6370 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
6371 ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
6375 case OFPACT_SET_L4_SRC_PORT:
6376 ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
6379 case OFPACT_SET_L4_DST_PORT:
6380 ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
6383 case OFPACT_RESUBMIT:
6384 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
6387 case OFPACT_SET_TUNNEL:
6388 ctx->flow.tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
6391 case OFPACT_SET_QUEUE:
6392 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
6395 case OFPACT_POP_QUEUE:
6396 ctx->flow.skb_priority = ctx->orig_skb_priority;
6399 case OFPACT_REG_MOVE:
6400 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow);
6403 case OFPACT_REG_LOAD:
6404 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow);
6407 case OFPACT_DEC_TTL:
6408 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6414 /* Nothing to do. */
6417 case OFPACT_MULTIPATH:
6418 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow);
6421 case OFPACT_AUTOPATH:
6422 xlate_autopath(ctx, ofpact_get_AUTOPATH(a));
6426 ctx->ofproto->has_bundle_action = true;
6427 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6430 case OFPACT_OUTPUT_REG:
6431 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6435 ctx->has_learn = true;
6436 if (ctx->may_learn) {
6437 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6445 case OFPACT_FIN_TIMEOUT:
6446 ctx->has_fin_timeout = true;
6447 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6450 case OFPACT_CLEAR_ACTIONS:
6452 * Nothing to do because writa-actions is not supported for now.
6453 * When writa-actions is supported, clear-actions also must
6454 * be supported at the same time.
6458 case OFPACT_WRITE_METADATA:
6459 metadata = ofpact_get_WRITE_METADATA(a);
6460 ctx->flow.metadata &= ~metadata->mask;
6461 ctx->flow.metadata |= metadata->metadata & metadata->mask;
6464 case OFPACT_GOTO_TABLE: {
6465 /* XXX remove recursion */
6466 /* It is assumed that goto-table is last action */
6467 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6468 ovs_assert(ctx->table_id < ogt->table_id);
6469 xlate_table_action(ctx, ctx->flow.in_port, ogt->table_id, true);
6477 ctx->rule->up.evictable = was_evictable;
6482 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
6483 struct ofproto_dpif *ofproto, const struct flow *flow,
6484 ovs_be16 initial_tci, struct rule_dpif *rule,
6485 uint8_t tcp_flags, const struct ofpbuf *packet)
6487 ovs_be64 initial_tun_id = flow->tunnel.tun_id;
6489 /* Flow initialization rules:
6490 * - 'base_flow' must match the kernel's view of the packet at the
6491 * time that action processing starts. 'flow' represents any
6492 * transformations we wish to make through actions.
6493 * - By default 'base_flow' and 'flow' are the same since the input
6494 * packet matches the output before any actions are applied.
6495 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
6496 * of the received packet as seen by the kernel. If we later output
6497 * to another device without any modifications this will cause us to
6498 * insert a new tag since the original one was stripped off by the
6500 * - Tunnel 'flow' is largely cleared when transitioning between
6501 * the input and output stages since it does not make sense to output
6502 * a packet with the exact headers that it was received with (i.e.
6503 * the destination IP is us). The one exception is the tun_id, which
6504 * is preserved to allow use in later resubmit lookups and loads into
6506 * - Tunnel 'base_flow' is completely cleared since that is what the
6507 * kernel does. If we wish to maintain the original values an action
6508 * needs to be generated. */
6510 ctx->ofproto = ofproto;
6512 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
6513 ctx->base_flow = ctx->flow;
6514 ctx->base_flow.vlan_tci = initial_tci;
6515 ctx->flow.tunnel.tun_id = initial_tun_id;
6517 ctx->packet = packet;
6518 ctx->may_learn = packet != NULL;
6519 ctx->tcp_flags = tcp_flags;
6520 ctx->resubmit_hook = NULL;
6521 ctx->report_hook = NULL;
6522 ctx->resubmit_stats = NULL;
6525 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
6526 * into datapath actions in 'odp_actions', using 'ctx'. */
6528 xlate_actions(struct action_xlate_ctx *ctx,
6529 const struct ofpact *ofpacts, size_t ofpacts_len,
6530 struct ofpbuf *odp_actions)
6532 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
6533 * that in the future we always keep a copy of the original flow for
6534 * tracing purposes. */
6535 static bool hit_resubmit_limit;
6537 enum slow_path_reason special;
6538 struct ofport_dpif *in_port;
6540 COVERAGE_INC(ofproto_dpif_xlate);
6542 ofpbuf_clear(odp_actions);
6543 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
6545 ctx->odp_actions = odp_actions;
6548 ctx->has_learn = false;
6549 ctx->has_normal = false;
6550 ctx->has_fin_timeout = false;
6551 ctx->nf_output_iface = NF_OUT_DROP;
6554 ctx->max_resubmit_trigger = false;
6555 ctx->orig_skb_priority = ctx->flow.skb_priority;
6559 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
6560 /* Do this conditionally because the copy is expensive enough that it
6561 * shows up in profiles.
6563 * We keep orig_flow in 'ctx' only because I couldn't make GCC 4.4
6564 * believe that I wasn't using it without initializing it if I kept it
6565 * in a local variable. */
6566 ctx->orig_flow = ctx->flow;
6569 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
6570 switch (ctx->ofproto->up.frag_handling) {
6571 case OFPC_FRAG_NORMAL:
6572 /* We must pretend that transport ports are unavailable. */
6573 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
6574 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
6577 case OFPC_FRAG_DROP:
6580 case OFPC_FRAG_REASM:
6583 case OFPC_FRAG_NX_MATCH:
6584 /* Nothing to do. */
6587 case OFPC_INVALID_TTL_TO_CONTROLLER:
6592 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
6593 special = process_special(ctx->ofproto, &ctx->flow, in_port, ctx->packet);
6595 ctx->slow |= special;
6597 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6598 ovs_be16 initial_tci = ctx->base_flow.vlan_tci;
6599 uint32_t local_odp_port;
6601 add_sflow_action(ctx);
6603 if (!in_port || may_receive(in_port, ctx)) {
6604 do_xlate_actions(ofpacts, ofpacts_len, ctx);
6606 /* We've let OFPP_NORMAL and the learning action look at the
6607 * packet, so drop it now if forwarding is disabled. */
6608 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
6609 ofpbuf_clear(ctx->odp_actions);
6610 add_sflow_action(ctx);
6614 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
6615 if (!hit_resubmit_limit) {
6616 /* We didn't record the original flow. Make sure we do from
6618 hit_resubmit_limit = true;
6619 } else if (!VLOG_DROP_ERR(&trace_rl)) {
6620 struct ds ds = DS_EMPTY_INITIALIZER;
6622 ofproto_trace(ctx->ofproto, &ctx->orig_flow, ctx->packet,
6624 VLOG_ERR("Trace triggered by excessive resubmit "
6625 "recursion:\n%s", ds_cstr(&ds));
6630 local_odp_port = ofp_port_to_odp_port(ctx->ofproto, OFPP_LOCAL);
6631 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
6633 ctx->odp_actions->data,
6634 ctx->odp_actions->size)) {
6635 ctx->slow |= SLOW_IN_BAND;
6637 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
6639 compose_output_action(ctx, OFPP_LOCAL);
6642 if (ctx->ofproto->has_mirrors) {
6643 add_mirror_actions(ctx, &ctx->orig_flow);
6645 fix_sflow_action(ctx);
6649 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
6650 * into datapath actions, using 'ctx', and discards the datapath actions. */
6652 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
6653 const struct ofpact *ofpacts,
6656 uint64_t odp_actions_stub[1024 / 8];
6657 struct ofpbuf odp_actions;
6659 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
6660 xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions);
6661 ofpbuf_uninit(&odp_actions);
6665 xlate_report(struct action_xlate_ctx *ctx, const char *s)
6667 if (ctx->report_hook) {
6668 ctx->report_hook(ctx, s);
6672 /* OFPP_NORMAL implementation. */
6674 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
6676 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
6677 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
6678 * the bundle on which the packet was received, returns the VLAN to which the
6681 * Both 'vid' and the return value are in the range 0...4095. */
6683 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
6685 switch (in_bundle->vlan_mode) {
6686 case PORT_VLAN_ACCESS:
6687 return in_bundle->vlan;
6690 case PORT_VLAN_TRUNK:
6693 case PORT_VLAN_NATIVE_UNTAGGED:
6694 case PORT_VLAN_NATIVE_TAGGED:
6695 return vid ? vid : in_bundle->vlan;
6702 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
6703 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
6706 * 'vid' should be the VID obtained from the 802.1Q header that was received as
6707 * part of a packet (specify 0 if there was no 802.1Q header), in the range
6710 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
6712 /* Allow any VID on the OFPP_NONE port. */
6713 if (in_bundle == &ofpp_none_bundle) {
6717 switch (in_bundle->vlan_mode) {
6718 case PORT_VLAN_ACCESS:
6721 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6722 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
6723 "packet received on port %s configured as VLAN "
6724 "%"PRIu16" access port",
6725 in_bundle->ofproto->up.name, vid,
6726 in_bundle->name, in_bundle->vlan);
6732 case PORT_VLAN_NATIVE_UNTAGGED:
6733 case PORT_VLAN_NATIVE_TAGGED:
6735 /* Port must always carry its native VLAN. */
6739 case PORT_VLAN_TRUNK:
6740 if (!ofbundle_includes_vlan(in_bundle, vid)) {
6742 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6743 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
6744 "received on port %s not configured for trunking "
6746 in_bundle->ofproto->up.name, vid,
6747 in_bundle->name, vid);
6759 /* Given 'vlan', the VLAN that a packet belongs to, and
6760 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
6761 * that should be included in the 802.1Q header. (If the return value is 0,
6762 * then the 802.1Q header should only be included in the packet if there is a
6765 * Both 'vlan' and the return value are in the range 0...4095. */
6767 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
6769 switch (out_bundle->vlan_mode) {
6770 case PORT_VLAN_ACCESS:
6773 case PORT_VLAN_TRUNK:
6774 case PORT_VLAN_NATIVE_TAGGED:
6777 case PORT_VLAN_NATIVE_UNTAGGED:
6778 return vlan == out_bundle->vlan ? 0 : vlan;
6786 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
6789 struct ofport_dpif *port;
6791 ovs_be16 tci, old_tci;
6793 vid = output_vlan_to_vid(out_bundle, vlan);
6794 if (!out_bundle->bond) {
6795 port = ofbundle_get_a_port(out_bundle);
6797 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
6800 /* No slaves enabled, so drop packet. */
6805 old_tci = ctx->flow.vlan_tci;
6807 if (tci || out_bundle->use_priority_tags) {
6808 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
6810 tci |= htons(VLAN_CFI);
6813 ctx->flow.vlan_tci = tci;
6815 compose_output_action(ctx, port->up.ofp_port);
6816 ctx->flow.vlan_tci = old_tci;
6820 mirror_mask_ffs(mirror_mask_t mask)
6822 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
6827 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
6829 return (bundle->vlan_mode != PORT_VLAN_ACCESS
6830 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
6834 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
6836 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
6839 /* Returns an arbitrary interface within 'bundle'. */
6840 static struct ofport_dpif *
6841 ofbundle_get_a_port(const struct ofbundle *bundle)
6843 return CONTAINER_OF(list_front(&bundle->ports),
6844 struct ofport_dpif, bundle_node);
6848 vlan_is_mirrored(const struct ofmirror *m, int vlan)
6850 return !m->vlans || bitmap_is_set(m->vlans, vlan);
6854 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
6856 struct ofproto_dpif *ofproto = ctx->ofproto;
6857 mirror_mask_t mirrors;
6858 struct ofbundle *in_bundle;
6861 const struct nlattr *a;
6864 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
6865 ctx->packet != NULL, NULL);
6869 mirrors = in_bundle->src_mirrors;
6871 /* Drop frames on bundles reserved for mirroring. */
6872 if (in_bundle->mirror_out) {
6873 if (ctx->packet != NULL) {
6874 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6875 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
6876 "%s, which is reserved exclusively for mirroring",
6877 ctx->ofproto->up.name, in_bundle->name);
6883 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
6884 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
6887 vlan = input_vid_to_vlan(in_bundle, vid);
6889 /* Look at the output ports to check for destination selections. */
6891 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
6892 ctx->odp_actions->size) {
6893 enum ovs_action_attr type = nl_attr_type(a);
6894 struct ofport_dpif *ofport;
6896 if (type != OVS_ACTION_ATTR_OUTPUT) {
6900 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
6901 if (ofport && ofport->bundle) {
6902 mirrors |= ofport->bundle->dst_mirrors;
6910 /* Restore the original packet before adding the mirror actions. */
6911 ctx->flow = *orig_flow;
6916 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6918 if (!vlan_is_mirrored(m, vlan)) {
6919 mirrors = zero_rightmost_1bit(mirrors);
6923 mirrors &= ~m->dup_mirrors;
6924 ctx->mirrors |= m->dup_mirrors;
6926 output_normal(ctx, m->out, vlan);
6927 } else if (vlan != m->out_vlan
6928 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
6929 struct ofbundle *bundle;
6931 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
6932 if (ofbundle_includes_vlan(bundle, m->out_vlan)
6933 && !bundle->mirror_out) {
6934 output_normal(ctx, bundle, m->out_vlan);
6942 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
6943 uint64_t packets, uint64_t bytes)
6949 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
6952 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6955 /* In normal circumstances 'm' will not be NULL. However,
6956 * if mirrors are reconfigured, we can temporarily get out
6957 * of sync in facet_revalidate(). We could "correct" the
6958 * mirror list before reaching here, but doing that would
6959 * not properly account the traffic stats we've currently
6960 * accumulated for previous mirror configuration. */
6964 m->packet_count += packets;
6965 m->byte_count += bytes;
6969 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
6970 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
6971 * indicate this; newer upstream kernels use gratuitous ARP requests. */
6973 is_gratuitous_arp(const struct flow *flow)
6975 return (flow->dl_type == htons(ETH_TYPE_ARP)
6976 && eth_addr_is_broadcast(flow->dl_dst)
6977 && (flow->nw_proto == ARP_OP_REPLY
6978 || (flow->nw_proto == ARP_OP_REQUEST
6979 && flow->nw_src == flow->nw_dst)));
6983 update_learning_table(struct ofproto_dpif *ofproto,
6984 const struct flow *flow, int vlan,
6985 struct ofbundle *in_bundle)
6987 struct mac_entry *mac;
6989 /* Don't learn the OFPP_NONE port. */
6990 if (in_bundle == &ofpp_none_bundle) {
6994 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
6998 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
6999 if (is_gratuitous_arp(flow)) {
7000 /* We don't want to learn from gratuitous ARP packets that are
7001 * reflected back over bond slaves so we lock the learning table. */
7002 if (!in_bundle->bond) {
7003 mac_entry_set_grat_arp_lock(mac);
7004 } else if (mac_entry_is_grat_arp_locked(mac)) {
7009 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
7010 /* The log messages here could actually be useful in debugging,
7011 * so keep the rate limit relatively high. */
7012 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
7013 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
7014 "on port %s in VLAN %d",
7015 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
7016 in_bundle->name, vlan);
7018 mac->port.p = in_bundle;
7019 tag_set_add(&ofproto->backer->revalidate_set,
7020 mac_learning_changed(ofproto->ml, mac));
7024 static struct ofbundle *
7025 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
7026 bool warn, struct ofport_dpif **in_ofportp)
7028 struct ofport_dpif *ofport;
7030 /* Find the port and bundle for the received packet. */
7031 ofport = get_ofp_port(ofproto, in_port);
7033 *in_ofportp = ofport;
7035 if (ofport && ofport->bundle) {
7036 return ofport->bundle;
7039 /* Special-case OFPP_NONE, which a controller may use as the ingress
7040 * port for traffic that it is sourcing. */
7041 if (in_port == OFPP_NONE) {
7042 return &ofpp_none_bundle;
7045 /* Odd. A few possible reasons here:
7047 * - We deleted a port but there are still a few packets queued up
7050 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
7051 * we don't know about.
7053 * - The ofproto client didn't configure the port as part of a bundle.
7054 * This is particularly likely to happen if a packet was received on the
7055 * port after it was created, but before the client had a chance to
7056 * configure its bundle.
7059 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7061 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
7062 "port %"PRIu16, ofproto->up.name, in_port);
7067 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
7068 * dropped. Returns true if they may be forwarded, false if they should be
7071 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
7072 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
7074 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
7075 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
7076 * checked by input_vid_is_valid().
7078 * May also add tags to '*tags', although the current implementation only does
7079 * so in one special case.
7082 is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port,
7085 struct ofproto_dpif *ofproto = ctx->ofproto;
7086 struct flow *flow = &ctx->flow;
7087 struct ofbundle *in_bundle = in_port->bundle;
7089 /* Drop frames for reserved multicast addresses
7090 * only if forward_bpdu option is absent. */
7091 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
7092 xlate_report(ctx, "packet has reserved destination MAC, dropping");
7096 if (in_bundle->bond) {
7097 struct mac_entry *mac;
7099 switch (bond_check_admissibility(in_bundle->bond, in_port,
7100 flow->dl_dst, &ctx->tags)) {
7105 xlate_report(ctx, "bonding refused admissibility, dropping");
7108 case BV_DROP_IF_MOVED:
7109 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
7110 if (mac && mac->port.p != in_bundle &&
7111 (!is_gratuitous_arp(flow)
7112 || mac_entry_is_grat_arp_locked(mac))) {
7113 xlate_report(ctx, "SLB bond thinks this packet looped back, "
7125 xlate_normal(struct action_xlate_ctx *ctx)
7127 struct ofport_dpif *in_port;
7128 struct ofbundle *in_bundle;
7129 struct mac_entry *mac;
7133 ctx->has_normal = true;
7135 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
7136 ctx->packet != NULL, &in_port);
7138 xlate_report(ctx, "no input bundle, dropping");
7142 /* Drop malformed frames. */
7143 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
7144 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
7145 if (ctx->packet != NULL) {
7146 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7147 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
7148 "VLAN tag received on port %s",
7149 ctx->ofproto->up.name, in_bundle->name);
7151 xlate_report(ctx, "partial VLAN tag, dropping");
7155 /* Drop frames on bundles reserved for mirroring. */
7156 if (in_bundle->mirror_out) {
7157 if (ctx->packet != NULL) {
7158 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7159 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7160 "%s, which is reserved exclusively for mirroring",
7161 ctx->ofproto->up.name, in_bundle->name);
7163 xlate_report(ctx, "input port is mirror output port, dropping");
7168 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
7169 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
7170 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
7173 vlan = input_vid_to_vlan(in_bundle, vid);
7175 /* Check other admissibility requirements. */
7176 if (in_port && !is_admissible(ctx, in_port, vlan)) {
7180 /* Learn source MAC. */
7181 if (ctx->may_learn) {
7182 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
7185 /* Determine output bundle. */
7186 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
7189 if (mac->port.p != in_bundle) {
7190 xlate_report(ctx, "forwarding to learned port");
7191 output_normal(ctx, mac->port.p, vlan);
7193 xlate_report(ctx, "learned port is input port, dropping");
7196 struct ofbundle *bundle;
7198 xlate_report(ctx, "no learned MAC for destination, flooding");
7199 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
7200 if (bundle != in_bundle
7201 && ofbundle_includes_vlan(bundle, vlan)
7202 && bundle->floodable
7203 && !bundle->mirror_out) {
7204 output_normal(ctx, bundle, vlan);
7207 ctx->nf_output_iface = NF_OUT_FLOOD;
7211 /* Optimized flow revalidation.
7213 * It's a difficult problem, in general, to tell which facets need to have
7214 * their actions recalculated whenever the OpenFlow flow table changes. We
7215 * don't try to solve that general problem: for most kinds of OpenFlow flow
7216 * table changes, we recalculate the actions for every facet. This is
7217 * relatively expensive, but it's good enough if the OpenFlow flow table
7218 * doesn't change very often.
7220 * However, we can expect one particular kind of OpenFlow flow table change to
7221 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
7222 * of CPU on revalidating every facet whenever MAC learning modifies the flow
7223 * table, we add a special case that applies to flow tables in which every rule
7224 * has the same form (that is, the same wildcards), except that the table is
7225 * also allowed to have a single "catch-all" flow that matches all packets. We
7226 * optimize this case by tagging all of the facets that resubmit into the table
7227 * and invalidating the same tag whenever a flow changes in that table. The
7228 * end result is that we revalidate just the facets that need it (and sometimes
7229 * a few more, but not all of the facets or even all of the facets that
7230 * resubmit to the table modified by MAC learning). */
7232 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
7233 * into an OpenFlow table with the given 'basis'. */
7235 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
7238 if (minimask_is_catchall(mask)) {
7241 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
7242 return tag_create_deterministic(hash);
7246 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
7247 * taggability of that table.
7249 * This function must be called after *each* change to a flow table. If you
7250 * skip calling it on some changes then the pointer comparisons at the end can
7251 * be invalid if you get unlucky. For example, if a flow removal causes a
7252 * cls_table to be destroyed and then a flow insertion causes a cls_table with
7253 * different wildcards to be created with the same address, then this function
7254 * will incorrectly skip revalidation. */
7256 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
7258 struct table_dpif *table = &ofproto->tables[table_id];
7259 const struct oftable *oftable = &ofproto->up.tables[table_id];
7260 struct cls_table *catchall, *other;
7261 struct cls_table *t;
7263 catchall = other = NULL;
7265 switch (hmap_count(&oftable->cls.tables)) {
7267 /* We could tag this OpenFlow table but it would make the logic a
7268 * little harder and it's a corner case that doesn't seem worth it
7274 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
7275 if (cls_table_is_catchall(t)) {
7277 } else if (!other) {
7280 /* Indicate that we can't tag this by setting both tables to
7281 * NULL. (We know that 'catchall' is already NULL.) */
7288 /* Can't tag this table. */
7292 if (table->catchall_table != catchall || table->other_table != other) {
7293 table->catchall_table = catchall;
7294 table->other_table = other;
7295 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7299 /* Given 'rule' that has changed in some way (either it is a rule being
7300 * inserted, a rule being deleted, or a rule whose actions are being
7301 * modified), marks facets for revalidation to ensure that packets will be
7302 * forwarded correctly according to the new state of the flow table.
7304 * This function must be called after *each* change to a flow table. See
7305 * the comment on table_update_taggable() for more information. */
7307 rule_invalidate(const struct rule_dpif *rule)
7309 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
7311 table_update_taggable(ofproto, rule->up.table_id);
7313 if (!ofproto->backer->need_revalidate) {
7314 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
7316 if (table->other_table && rule->tag) {
7317 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
7319 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7325 set_frag_handling(struct ofproto *ofproto_,
7326 enum ofp_config_flags frag_handling)
7328 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7329 if (frag_handling != OFPC_FRAG_REASM) {
7330 ofproto->backer->need_revalidate = REV_RECONFIGURE;
7338 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
7339 const struct flow *flow,
7340 const struct ofpact *ofpacts, size_t ofpacts_len)
7342 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7343 struct odputil_keybuf keybuf;
7344 struct dpif_flow_stats stats;
7348 struct action_xlate_ctx ctx;
7349 uint64_t odp_actions_stub[1024 / 8];
7350 struct ofpbuf odp_actions;
7352 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
7353 odp_flow_key_from_flow(&key, flow,
7354 ofp_port_to_odp_port(ofproto, flow->in_port));
7356 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
7358 action_xlate_ctx_init(&ctx, ofproto, flow, flow->vlan_tci, NULL,
7359 packet_get_tcp_flags(packet, flow), packet);
7360 ctx.resubmit_stats = &stats;
7362 ofpbuf_use_stub(&odp_actions,
7363 odp_actions_stub, sizeof odp_actions_stub);
7364 xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions);
7365 dpif_execute(ofproto->backer->dpif, key.data, key.size,
7366 odp_actions.data, odp_actions.size, packet);
7367 ofpbuf_uninit(&odp_actions);
7375 set_netflow(struct ofproto *ofproto_,
7376 const struct netflow_options *netflow_options)
7378 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7380 if (netflow_options) {
7381 if (!ofproto->netflow) {
7382 ofproto->netflow = netflow_create();
7384 return netflow_set_options(ofproto->netflow, netflow_options);
7386 netflow_destroy(ofproto->netflow);
7387 ofproto->netflow = NULL;
7393 get_netflow_ids(const struct ofproto *ofproto_,
7394 uint8_t *engine_type, uint8_t *engine_id)
7396 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7398 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
7402 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
7404 if (!facet_is_controller_flow(facet) &&
7405 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
7406 struct subfacet *subfacet;
7407 struct ofexpired expired;
7409 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
7410 if (subfacet->path == SF_FAST_PATH) {
7411 struct dpif_flow_stats stats;
7413 subfacet_reinstall(subfacet, &stats);
7414 subfacet_update_stats(subfacet, &stats);
7418 expired.flow = facet->flow;
7419 expired.packet_count = facet->packet_count;
7420 expired.byte_count = facet->byte_count;
7421 expired.used = facet->used;
7422 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
7427 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
7429 struct facet *facet;
7431 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7432 send_active_timeout(ofproto, facet);
7436 static struct ofproto_dpif *
7437 ofproto_dpif_lookup(const char *name)
7439 struct ofproto_dpif *ofproto;
7441 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
7442 hash_string(name, 0), &all_ofproto_dpifs) {
7443 if (!strcmp(ofproto->up.name, name)) {
7451 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
7452 const char *argv[], void *aux OVS_UNUSED)
7454 struct ofproto_dpif *ofproto;
7457 ofproto = ofproto_dpif_lookup(argv[1]);
7459 unixctl_command_reply_error(conn, "no such bridge");
7462 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7464 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7465 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7469 unixctl_command_reply(conn, "table successfully flushed");
7473 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
7474 const char *argv[], void *aux OVS_UNUSED)
7476 struct ds ds = DS_EMPTY_INITIALIZER;
7477 const struct ofproto_dpif *ofproto;
7478 const struct mac_entry *e;
7480 ofproto = ofproto_dpif_lookup(argv[1]);
7482 unixctl_command_reply_error(conn, "no such bridge");
7486 ds_put_cstr(&ds, " port VLAN MAC Age\n");
7487 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
7488 struct ofbundle *bundle = e->port.p;
7489 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
7490 ofbundle_get_a_port(bundle)->odp_port,
7491 e->vlan, ETH_ADDR_ARGS(e->mac),
7492 mac_entry_age(ofproto->ml, e));
7494 unixctl_command_reply(conn, ds_cstr(&ds));
7499 struct action_xlate_ctx ctx;
7505 trace_format_rule(struct ds *result, uint8_t table_id, int level,
7506 const struct rule_dpif *rule)
7508 ds_put_char_multiple(result, '\t', level);
7510 ds_put_cstr(result, "No match\n");
7514 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
7515 table_id, ntohll(rule->up.flow_cookie));
7516 cls_rule_format(&rule->up.cr, result);
7517 ds_put_char(result, '\n');
7519 ds_put_char_multiple(result, '\t', level);
7520 ds_put_cstr(result, "OpenFlow ");
7521 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
7522 ds_put_char(result, '\n');
7526 trace_format_flow(struct ds *result, int level, const char *title,
7527 struct trace_ctx *trace)
7529 ds_put_char_multiple(result, '\t', level);
7530 ds_put_format(result, "%s: ", title);
7531 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
7532 ds_put_cstr(result, "unchanged");
7534 flow_format(result, &trace->ctx.flow);
7535 trace->flow = trace->ctx.flow;
7537 ds_put_char(result, '\n');
7541 trace_format_regs(struct ds *result, int level, const char *title,
7542 struct trace_ctx *trace)
7546 ds_put_char_multiple(result, '\t', level);
7547 ds_put_format(result, "%s:", title);
7548 for (i = 0; i < FLOW_N_REGS; i++) {
7549 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
7551 ds_put_char(result, '\n');
7555 trace_format_odp(struct ds *result, int level, const char *title,
7556 struct trace_ctx *trace)
7558 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
7560 ds_put_char_multiple(result, '\t', level);
7561 ds_put_format(result, "%s: ", title);
7562 format_odp_actions(result, odp_actions->data, odp_actions->size);
7563 ds_put_char(result, '\n');
7567 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
7569 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7570 struct ds *result = trace->result;
7572 ds_put_char(result, '\n');
7573 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
7574 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
7575 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
7576 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
7580 trace_report(struct action_xlate_ctx *ctx, const char *s)
7582 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7583 struct ds *result = trace->result;
7585 ds_put_char_multiple(result, '\t', ctx->recurse);
7586 ds_put_cstr(result, s);
7587 ds_put_char(result, '\n');
7591 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
7592 void *aux OVS_UNUSED)
7594 const char *dpname = argv[1];
7595 struct ofproto_dpif *ofproto;
7596 struct ofpbuf odp_key;
7597 struct ofpbuf *packet;
7598 ovs_be16 initial_tci;
7604 ofpbuf_init(&odp_key, 0);
7607 ofproto = ofproto_dpif_lookup(dpname);
7609 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
7613 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
7614 /* ofproto/trace dpname flow [-generate] */
7615 const char *flow_s = argv[2];
7616 const char *generate_s = argv[3];
7618 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
7619 * flow. We guess which type it is based on whether 'flow_s' contains
7620 * an '(', since a datapath flow always contains '(') but an
7621 * OpenFlow-like flow should not (in fact it's allowed but I believe
7622 * that's not documented anywhere).
7624 * An alternative would be to try to parse 'flow_s' both ways, but then
7625 * it would be tricky giving a sensible error message. After all, do
7626 * you just say "syntax error" or do you present both error messages?
7627 * Both choices seem lousy. */
7628 if (strchr(flow_s, '(')) {
7631 /* Convert string to datapath key. */
7632 ofpbuf_init(&odp_key, 0);
7633 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
7635 unixctl_command_reply_error(conn, "Bad flow syntax");
7639 /* XXX: Since we allow the user to specify an ofproto, it's
7640 * possible they will specify a different ofproto than the one the
7641 * port actually belongs too. Ideally we should simply remove the
7642 * ability to specify the ofproto. */
7643 if (ofproto_receive(ofproto->backer, NULL, odp_key.data,
7644 odp_key.size, &flow, NULL, NULL, NULL,
7646 unixctl_command_reply_error(conn, "Invalid flow");
7652 error_s = parse_ofp_exact_flow(&flow, argv[2]);
7654 unixctl_command_reply_error(conn, error_s);
7659 initial_tci = flow.vlan_tci;
7662 /* Generate a packet, if requested. */
7664 packet = ofpbuf_new(0);
7665 flow_compose(packet, &flow);
7667 } else if (argc == 7) {
7668 /* ofproto/trace dpname priority tun_id in_port mark packet */
7669 const char *priority_s = argv[2];
7670 const char *tun_id_s = argv[3];
7671 const char *in_port_s = argv[4];
7672 const char *mark_s = argv[5];
7673 const char *packet_s = argv[6];
7674 uint32_t in_port = atoi(in_port_s);
7675 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
7676 uint32_t priority = atoi(priority_s);
7677 uint32_t mark = atoi(mark_s);
7680 msg = eth_from_hex(packet_s, &packet);
7682 unixctl_command_reply_error(conn, msg);
7686 ds_put_cstr(&result, "Packet: ");
7687 s = ofp_packet_to_string(packet->data, packet->size);
7688 ds_put_cstr(&result, s);
7691 flow_extract(packet, priority, mark, NULL, in_port, &flow);
7692 flow.tunnel.tun_id = tun_id;
7693 initial_tci = flow.vlan_tci;
7695 unixctl_command_reply_error(conn, "Bad command syntax");
7699 ofproto_trace(ofproto, &flow, packet, initial_tci, &result);
7700 unixctl_command_reply(conn, ds_cstr(&result));
7703 ds_destroy(&result);
7704 ofpbuf_delete(packet);
7705 ofpbuf_uninit(&odp_key);
7709 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
7710 const struct ofpbuf *packet, ovs_be16 initial_tci,
7713 struct rule_dpif *rule;
7715 ds_put_cstr(ds, "Flow: ");
7716 flow_format(ds, flow);
7717 ds_put_char(ds, '\n');
7719 rule = rule_dpif_lookup(ofproto, flow);
7721 trace_format_rule(ds, 0, 0, rule);
7722 if (rule == ofproto->miss_rule) {
7723 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
7724 } else if (rule == ofproto->no_packet_in_rule) {
7725 ds_put_cstr(ds, "\nNo match, packets dropped because "
7726 "OFPPC_NO_PACKET_IN is set on in_port.\n");
7730 uint64_t odp_actions_stub[1024 / 8];
7731 struct ofpbuf odp_actions;
7733 struct trace_ctx trace;
7736 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
7739 ofpbuf_use_stub(&odp_actions,
7740 odp_actions_stub, sizeof odp_actions_stub);
7741 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_tci,
7742 rule, tcp_flags, packet);
7743 trace.ctx.resubmit_hook = trace_resubmit;
7744 trace.ctx.report_hook = trace_report;
7745 xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len,
7748 ds_put_char(ds, '\n');
7749 trace_format_flow(ds, 0, "Final flow", &trace);
7750 ds_put_cstr(ds, "Datapath actions: ");
7751 format_odp_actions(ds, odp_actions.data, odp_actions.size);
7752 ofpbuf_uninit(&odp_actions);
7754 if (trace.ctx.slow) {
7755 enum slow_path_reason slow;
7757 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
7758 "slow path because it:");
7759 for (slow = trace.ctx.slow; slow; ) {
7760 enum slow_path_reason bit = rightmost_1bit(slow);
7764 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
7767 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
7770 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
7773 ds_put_cstr(ds, "\n\t- Needs in-band special case "
7776 ds_put_cstr(ds, "\n\t (The datapath actions are "
7777 "incomplete--for complete actions, "
7778 "please supply a packet.)");
7781 case SLOW_CONTROLLER:
7782 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
7783 "to the OpenFlow controller.");
7786 ds_put_cstr(ds, "\n\t- Needs more specific matching "
7787 "than the datapath supports.");
7794 if (slow & ~SLOW_MATCH) {
7795 ds_put_cstr(ds, "\nThe datapath actions above do not reflect "
7796 "the special slow-path processing.");
7803 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7804 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7807 unixctl_command_reply(conn, NULL);
7811 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7812 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7815 unixctl_command_reply(conn, NULL);
7818 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
7819 * 'reply' describing the results. */
7821 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
7823 struct facet *facet;
7827 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7828 if (!facet_check_consistency(facet)) {
7833 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
7837 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
7838 ofproto->up.name, errors);
7840 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
7845 ofproto_dpif_self_check(struct unixctl_conn *conn,
7846 int argc, const char *argv[], void *aux OVS_UNUSED)
7848 struct ds reply = DS_EMPTY_INITIALIZER;
7849 struct ofproto_dpif *ofproto;
7852 ofproto = ofproto_dpif_lookup(argv[1]);
7854 unixctl_command_reply_error(conn, "Unknown ofproto (use "
7855 "ofproto/list for help)");
7858 ofproto_dpif_self_check__(ofproto, &reply);
7860 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7861 ofproto_dpif_self_check__(ofproto, &reply);
7865 unixctl_command_reply(conn, ds_cstr(&reply));
7869 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
7870 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
7871 * to destroy 'ofproto_shash' and free the returned value. */
7872 static const struct shash_node **
7873 get_ofprotos(struct shash *ofproto_shash)
7875 const struct ofproto_dpif *ofproto;
7877 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7878 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
7879 shash_add_nocopy(ofproto_shash, name, ofproto);
7882 return shash_sort(ofproto_shash);
7886 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
7887 const char *argv[] OVS_UNUSED,
7888 void *aux OVS_UNUSED)
7890 struct ds ds = DS_EMPTY_INITIALIZER;
7891 struct shash ofproto_shash;
7892 const struct shash_node **sorted_ofprotos;
7895 shash_init(&ofproto_shash);
7896 sorted_ofprotos = get_ofprotos(&ofproto_shash);
7897 for (i = 0; i < shash_count(&ofproto_shash); i++) {
7898 const struct shash_node *node = sorted_ofprotos[i];
7899 ds_put_format(&ds, "%s\n", node->name);
7902 shash_destroy(&ofproto_shash);
7903 free(sorted_ofprotos);
7905 unixctl_command_reply(conn, ds_cstr(&ds));
7910 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
7912 struct dpif_dp_stats s;
7913 const struct shash_node **ports;
7916 dpif_get_dp_stats(ofproto->backer->dpif, &s);
7918 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
7919 dpif_name(ofproto->backer->dpif));
7920 /* xxx It would be better to show bridge-specific stats instead
7921 * xxx of dp ones. */
7923 "\tlookups: hit:%"PRIu64" missed:%"PRIu64" lost:%"PRIu64"\n",
7924 s.n_hit, s.n_missed, s.n_lost);
7925 ds_put_format(ds, "\tflows: %zu\n",
7926 hmap_count(&ofproto->subfacets));
7928 ports = shash_sort(&ofproto->up.port_by_name);
7929 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
7930 const struct shash_node *node = ports[i];
7931 struct ofport *ofport = node->data;
7932 const char *name = netdev_get_name(ofport->netdev);
7933 const char *type = netdev_get_type(ofport->netdev);
7936 ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port);
7938 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
7939 if (odp_port != OVSP_NONE) {
7940 ds_put_format(ds, "%"PRIu32":", odp_port);
7942 ds_put_cstr(ds, "none:");
7945 if (strcmp(type, "system")) {
7946 struct netdev *netdev;
7949 ds_put_format(ds, " (%s", type);
7951 error = netdev_open(name, type, &netdev);
7956 error = netdev_get_config(netdev, &config);
7958 const struct smap_node **nodes;
7961 nodes = smap_sort(&config);
7962 for (i = 0; i < smap_count(&config); i++) {
7963 const struct smap_node *node = nodes[i];
7964 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
7965 node->key, node->value);
7969 smap_destroy(&config);
7971 netdev_close(netdev);
7973 ds_put_char(ds, ')');
7975 ds_put_char(ds, '\n');
7981 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
7982 const char *argv[], void *aux OVS_UNUSED)
7984 struct ds ds = DS_EMPTY_INITIALIZER;
7985 const struct ofproto_dpif *ofproto;
7989 for (i = 1; i < argc; i++) {
7990 ofproto = ofproto_dpif_lookup(argv[i]);
7992 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
7993 "for help)", argv[i]);
7994 unixctl_command_reply_error(conn, ds_cstr(&ds));
7997 show_dp_format(ofproto, &ds);
8000 struct shash ofproto_shash;
8001 const struct shash_node **sorted_ofprotos;
8004 shash_init(&ofproto_shash);
8005 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8006 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8007 const struct shash_node *node = sorted_ofprotos[i];
8008 show_dp_format(node->data, &ds);
8011 shash_destroy(&ofproto_shash);
8012 free(sorted_ofprotos);
8015 unixctl_command_reply(conn, ds_cstr(&ds));
8020 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
8021 int argc OVS_UNUSED, const char *argv[],
8022 void *aux OVS_UNUSED)
8024 struct ds ds = DS_EMPTY_INITIALIZER;
8025 const struct ofproto_dpif *ofproto;
8026 struct subfacet *subfacet;
8028 ofproto = ofproto_dpif_lookup(argv[1]);
8030 unixctl_command_reply_error(conn, "no such bridge");
8034 update_stats(ofproto->backer);
8036 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
8037 struct odputil_keybuf keybuf;
8040 subfacet_get_key(subfacet, &keybuf, &key);
8041 odp_flow_key_format(key.data, key.size, &ds);
8043 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
8044 subfacet->dp_packet_count, subfacet->dp_byte_count);
8045 if (subfacet->used) {
8046 ds_put_format(&ds, "%.3fs",
8047 (time_msec() - subfacet->used) / 1000.0);
8049 ds_put_format(&ds, "never");
8051 if (subfacet->facet->tcp_flags) {
8052 ds_put_cstr(&ds, ", flags:");
8053 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
8056 ds_put_cstr(&ds, ", actions:");
8057 format_odp_actions(&ds, subfacet->actions, subfacet->actions_len);
8058 ds_put_char(&ds, '\n');
8061 unixctl_command_reply(conn, ds_cstr(&ds));
8066 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
8067 int argc OVS_UNUSED, const char *argv[],
8068 void *aux OVS_UNUSED)
8070 struct ds ds = DS_EMPTY_INITIALIZER;
8071 struct ofproto_dpif *ofproto;
8073 ofproto = ofproto_dpif_lookup(argv[1]);
8075 unixctl_command_reply_error(conn, "no such bridge");
8079 flush(&ofproto->up);
8081 unixctl_command_reply(conn, ds_cstr(&ds));
8086 ofproto_dpif_unixctl_init(void)
8088 static bool registered;
8094 unixctl_command_register(
8096 "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}",
8097 2, 6, ofproto_unixctl_trace, NULL);
8098 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
8099 ofproto_unixctl_fdb_flush, NULL);
8100 unixctl_command_register("fdb/show", "bridge", 1, 1,
8101 ofproto_unixctl_fdb_show, NULL);
8102 unixctl_command_register("ofproto/clog", "", 0, 0,
8103 ofproto_dpif_clog, NULL);
8104 unixctl_command_register("ofproto/unclog", "", 0, 0,
8105 ofproto_dpif_unclog, NULL);
8106 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
8107 ofproto_dpif_self_check, NULL);
8108 unixctl_command_register("dpif/dump-dps", "", 0, 0,
8109 ofproto_unixctl_dpif_dump_dps, NULL);
8110 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
8111 ofproto_unixctl_dpif_show, NULL);
8112 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
8113 ofproto_unixctl_dpif_dump_flows, NULL);
8114 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
8115 ofproto_unixctl_dpif_del_flows, NULL);
8118 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
8120 * This is deprecated. It is only for compatibility with broken device drivers
8121 * in old versions of Linux that do not properly support VLANs when VLAN
8122 * devices are not used. When broken device drivers are no longer in
8123 * widespread use, we will delete these interfaces. */
8126 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
8128 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
8129 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
8131 if (realdev_ofp_port == ofport->realdev_ofp_port
8132 && vid == ofport->vlandev_vid) {
8136 ofproto->backer->need_revalidate = REV_RECONFIGURE;
8138 if (ofport->realdev_ofp_port) {
8141 if (realdev_ofp_port && ofport->bundle) {
8142 /* vlandevs are enslaved to their realdevs, so they are not allowed to
8143 * themselves be part of a bundle. */
8144 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
8147 ofport->realdev_ofp_port = realdev_ofp_port;
8148 ofport->vlandev_vid = vid;
8150 if (realdev_ofp_port) {
8151 vsp_add(ofport, realdev_ofp_port, vid);
8158 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
8160 return hash_2words(realdev_ofp_port, vid);
8163 /* Returns the ODP port number of the Linux VLAN device that corresponds to
8164 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
8165 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
8166 * it would return the port number of eth0.9.
8168 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
8169 * function just returns its 'realdev_odp_port' argument. */
8171 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
8172 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
8174 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
8175 uint16_t realdev_ofp_port;
8176 int vid = vlan_tci_to_vid(vlan_tci);
8177 const struct vlan_splinter *vsp;
8179 realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port);
8180 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
8181 hash_realdev_vid(realdev_ofp_port, vid),
8182 &ofproto->realdev_vid_map) {
8183 if (vsp->realdev_ofp_port == realdev_ofp_port
8184 && vsp->vid == vid) {
8185 return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port);
8189 return realdev_odp_port;
8192 static struct vlan_splinter *
8193 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
8195 struct vlan_splinter *vsp;
8197 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
8198 &ofproto->vlandev_map) {
8199 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
8207 /* Returns the OpenFlow port number of the "real" device underlying the Linux
8208 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
8209 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
8210 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
8211 * eth0 and store 9 in '*vid'.
8213 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
8214 * VLAN device. Unless VLAN splinters are enabled, this is what this function
8217 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
8218 uint16_t vlandev_ofp_port, int *vid)
8220 if (!hmap_is_empty(&ofproto->vlandev_map)) {
8221 const struct vlan_splinter *vsp;
8223 vsp = vlandev_find(ofproto, vlandev_ofp_port);
8228 return vsp->realdev_ofp_port;
8234 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
8235 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
8236 * 'flow->in_port' to the "real" device backing the VLAN device, sets
8237 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
8238 * always the case unless VLAN splinters are enabled), returns false without
8239 * making any changes. */
8241 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
8246 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
8251 /* Cause the flow to be processed as if it came in on the real device with
8252 * the VLAN device's VLAN ID. */
8253 flow->in_port = realdev;
8254 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
8259 vsp_remove(struct ofport_dpif *port)
8261 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8262 struct vlan_splinter *vsp;
8264 vsp = vlandev_find(ofproto, port->up.ofp_port);
8266 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
8267 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
8270 port->realdev_ofp_port = 0;
8272 VLOG_ERR("missing vlan device record");
8277 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
8279 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8281 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
8282 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
8283 == realdev_ofp_port)) {
8284 struct vlan_splinter *vsp;
8286 vsp = xmalloc(sizeof *vsp);
8287 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
8288 hash_int(port->up.ofp_port, 0));
8289 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
8290 hash_realdev_vid(realdev_ofp_port, vid));
8291 vsp->realdev_ofp_port = realdev_ofp_port;
8292 vsp->vlandev_ofp_port = port->up.ofp_port;
8295 port->realdev_ofp_port = realdev_ofp_port;
8297 VLOG_ERR("duplicate vlan device record");
8302 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
8304 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
8305 return ofport ? ofport->odp_port : OVSP_NONE;
8308 static struct ofport_dpif *
8309 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
8311 struct ofport_dpif *port;
8313 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
8314 hash_int(odp_port, 0),
8315 &backer->odp_to_ofport_map) {
8316 if (port->odp_port == odp_port) {
8325 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
8327 struct ofport_dpif *port;
8329 port = odp_port_to_ofport(ofproto->backer, odp_port);
8330 if (port && &ofproto->up == port->up.ofproto) {
8331 return port->up.ofp_port;
8337 const struct ofproto_class ofproto_dpif_class = {
8372 port_is_lacp_current,
8373 NULL, /* rule_choose_table */
8380 rule_modify_actions,
8389 get_cfm_remote_mpids,
8394 get_stp_port_status,
8401 is_mirror_output_bundle,
8402 forward_bpdu_changed,
8403 set_mac_table_config,