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"
25 #include "byte-order.h"
30 #include "dynamic-string.h"
31 #include "fail-open.h"
35 #include "mac-learning.h"
36 #include "meta-flow.h"
37 #include "multipath.h"
38 #include "netdev-vport.h"
45 #include "ofp-actions.h"
46 #include "ofp-parse.h"
47 #include "ofp-print.h"
48 #include "ofproto-dpif-governor.h"
49 #include "ofproto-dpif-sflow.h"
50 #include "poll-loop.h"
55 #include "unaligned.h"
57 #include "vlan-bitmap.h"
60 VLOG_DEFINE_THIS_MODULE(ofproto_dpif);
62 COVERAGE_DEFINE(ofproto_dpif_expired);
63 COVERAGE_DEFINE(ofproto_dpif_xlate);
64 COVERAGE_DEFINE(facet_changed_rule);
65 COVERAGE_DEFINE(facet_revalidate);
66 COVERAGE_DEFINE(facet_unexpected);
67 COVERAGE_DEFINE(facet_suppress);
69 /* Maximum depth of flow table recursion (due to resubmit actions) in a
70 * flow translation. */
71 #define MAX_RESUBMIT_RECURSION 64
73 /* Number of implemented OpenFlow tables. */
74 enum { N_TABLES = 255 };
75 enum { TBL_INTERNAL = N_TABLES - 1 }; /* Used for internal hidden rules. */
76 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 facet *, const struct dpif_flow_stats *);
123 static tag_type rule_calculate_tag(const struct flow *,
124 const struct minimask *, uint32_t basis);
125 static void rule_invalidate(const struct rule_dpif *);
127 #define MAX_MIRRORS 32
128 typedef uint32_t mirror_mask_t;
129 #define MIRROR_MASK_C(X) UINT32_C(X)
130 BUILD_ASSERT_DECL(sizeof(mirror_mask_t) * CHAR_BIT >= MAX_MIRRORS);
132 struct ofproto_dpif *ofproto; /* Owning ofproto. */
133 size_t idx; /* In ofproto's "mirrors" array. */
134 void *aux; /* Key supplied by ofproto's client. */
135 char *name; /* Identifier for log messages. */
137 /* Selection criteria. */
138 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
139 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
140 unsigned long *vlans; /* Bitmap of chosen VLANs, NULL selects all. */
142 /* Output (exactly one of out == NULL and out_vlan == -1 is true). */
143 struct ofbundle *out; /* Output port or NULL. */
144 int out_vlan; /* Output VLAN or -1. */
145 mirror_mask_t dup_mirrors; /* Bitmap of mirrors with the same output. */
148 int64_t packet_count; /* Number of packets sent. */
149 int64_t byte_count; /* Number of bytes sent. */
152 static void mirror_destroy(struct ofmirror *);
153 static void update_mirror_stats(struct ofproto_dpif *ofproto,
154 mirror_mask_t mirrors,
155 uint64_t packets, uint64_t bytes);
158 struct hmap_node hmap_node; /* In struct ofproto's "bundles" hmap. */
159 struct ofproto_dpif *ofproto; /* Owning ofproto. */
160 void *aux; /* Key supplied by ofproto's client. */
161 char *name; /* Identifier for log messages. */
164 struct list ports; /* Contains "struct ofport"s. */
165 enum port_vlan_mode vlan_mode; /* VLAN mode */
166 int vlan; /* -1=trunk port, else a 12-bit VLAN ID. */
167 unsigned long *trunks; /* Bitmap of trunked VLANs, if 'vlan' == -1.
168 * NULL if all VLANs are trunked. */
169 struct lacp *lacp; /* LACP if LACP is enabled, otherwise NULL. */
170 struct bond *bond; /* Nonnull iff more than one port. */
171 bool use_priority_tags; /* Use 802.1p tag for frames in VLAN 0? */
174 bool floodable; /* True if no port has OFPUTIL_PC_NO_FLOOD set. */
176 /* Port mirroring info. */
177 mirror_mask_t src_mirrors; /* Mirrors triggered when packet received. */
178 mirror_mask_t dst_mirrors; /* Mirrors triggered when packet sent. */
179 mirror_mask_t mirror_out; /* Mirrors that output to this bundle. */
182 static void bundle_remove(struct ofport *);
183 static void bundle_update(struct ofbundle *);
184 static void bundle_destroy(struct ofbundle *);
185 static void bundle_del_port(struct ofport_dpif *);
186 static void bundle_run(struct ofbundle *);
187 static void bundle_wait(struct ofbundle *);
188 static struct ofbundle *lookup_input_bundle(const struct ofproto_dpif *,
189 uint16_t in_port, bool warn,
190 struct ofport_dpif **in_ofportp);
192 /* A controller may use OFPP_NONE as the ingress port to indicate that
193 * it did not arrive on a "real" port. 'ofpp_none_bundle' exists for
194 * when an input bundle is needed for validation (e.g., mirroring or
195 * OFPP_NORMAL processing). It is not connected to an 'ofproto' or have
196 * any 'port' structs, so care must be taken when dealing with it. */
197 static struct ofbundle ofpp_none_bundle = {
199 .vlan_mode = PORT_VLAN_TRUNK
202 static void stp_run(struct ofproto_dpif *ofproto);
203 static void stp_wait(struct ofproto_dpif *ofproto);
204 static int set_stp_port(struct ofport *,
205 const struct ofproto_port_stp_settings *);
207 static bool ofbundle_includes_vlan(const struct ofbundle *, uint16_t vlan);
209 struct action_xlate_ctx {
210 /* action_xlate_ctx_init() initializes these members. */
213 struct ofproto_dpif *ofproto;
215 /* Flow to which the OpenFlow actions apply. xlate_actions() will modify
216 * this flow when actions change header fields. */
219 /* stack for the push and pop actions.
220 * Each stack element is of the type "union mf_subvalue". */
222 union mf_subvalue init_stack[1024 / sizeof(union mf_subvalue)];
224 /* The packet corresponding to 'flow', or a null pointer if we are
225 * revalidating without a packet to refer to. */
226 const struct ofpbuf *packet;
228 /* Should OFPP_NORMAL update the MAC learning table? Should "learn"
229 * actions update the flow table?
231 * We want to update these tables if we are actually processing a packet,
232 * or if we are accounting for packets that the datapath has processed, but
233 * not if we are just revalidating. */
236 /* The rule that we are currently translating, or NULL. */
237 struct rule_dpif *rule;
239 /* Union of the set of TCP flags seen so far in this flow. (Used only by
240 * NXAST_FIN_TIMEOUT. Set to zero to avoid updating updating rules'
244 /* If nonnull, flow translation calls this function just before executing a
245 * resubmit or OFPP_TABLE action. In addition, disables logging of traces
246 * when the recursion depth is exceeded.
248 * 'rule' is the rule being submitted into. It will be null if the
249 * resubmit or OFPP_TABLE action didn't find a matching rule.
251 * This is normally null so the client has to set it manually after
252 * calling action_xlate_ctx_init(). */
253 void (*resubmit_hook)(struct action_xlate_ctx *, struct rule_dpif *rule);
255 /* If nonnull, flow translation calls this function to report some
256 * significant decision, e.g. to explain why OFPP_NORMAL translation
257 * dropped a packet. */
258 void (*report_hook)(struct action_xlate_ctx *, const char *s);
260 /* If nonnull, flow translation credits the specified statistics to each
261 * rule reached through a resubmit or OFPP_TABLE action.
263 * This is normally null so the client has to set it manually after
264 * calling action_xlate_ctx_init(). */
265 const struct dpif_flow_stats *resubmit_stats;
267 /* xlate_actions() initializes and uses these members. The client might want
268 * to look at them after it returns. */
270 struct ofpbuf *odp_actions; /* Datapath actions. */
271 tag_type tags; /* Tags associated with actions. */
272 enum slow_path_reason slow; /* 0 if fast path may be used. */
273 bool has_learn; /* Actions include NXAST_LEARN? */
274 bool has_normal; /* Actions output to OFPP_NORMAL? */
275 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
276 uint16_t nf_output_iface; /* Output interface index for NetFlow. */
277 mirror_mask_t mirrors; /* Bitmap of associated mirrors. */
279 /* xlate_actions() initializes and uses these members, but the client has no
280 * reason to look at them. */
282 int recurse; /* Recursion level, via xlate_table_action. */
283 bool max_resubmit_trigger; /* Recursed too deeply during translation. */
284 struct flow base_flow; /* Flow at the last commit. */
285 uint32_t orig_skb_priority; /* Priority when packet arrived. */
286 uint8_t table_id; /* OpenFlow table ID where flow was found. */
287 uint32_t sflow_n_outputs; /* Number of output ports. */
288 uint32_t sflow_odp_port; /* Output port for composing sFlow action. */
289 uint16_t user_cookie_offset;/* Used for user_action_cookie fixup. */
290 bool exit; /* No further actions should be processed. */
293 /* Initial values of fields of the packet that may be changed during
294 * flow processing and needed later. */
295 struct initial_vals {
296 /* This is the value of vlan_tci in the packet as actually received from
297 * dpif. This is the same as the facet's flow.vlan_tci unless the packet
298 * was received via a VLAN splinter. In that case, this value is 0
299 * (because the packet as actually received from the dpif had no 802.1Q
300 * tag) but the facet's flow.vlan_tci is set to the VLAN that the splinter
303 * This member should be removed when the VLAN splinters feature is no
307 /* If received on a tunnel, the IP TOS value of the tunnel. */
308 uint8_t tunnel_ip_tos;
311 static void action_xlate_ctx_init(struct action_xlate_ctx *,
312 struct ofproto_dpif *, const struct flow *,
313 const struct initial_vals *initial_vals,
315 uint8_t tcp_flags, const struct ofpbuf *);
316 static void xlate_actions(struct action_xlate_ctx *,
317 const struct ofpact *ofpacts, size_t ofpacts_len,
318 struct ofpbuf *odp_actions);
319 static void xlate_actions_for_side_effects(struct action_xlate_ctx *,
320 const struct ofpact *ofpacts,
322 static void xlate_table_action(struct action_xlate_ctx *, uint16_t in_port,
323 uint8_t table_id, bool may_packet_in);
325 static size_t put_userspace_action(const struct ofproto_dpif *,
326 struct ofpbuf *odp_actions,
328 const union user_action_cookie *);
330 static void compose_slow_path(const struct ofproto_dpif *, const struct flow *,
331 enum slow_path_reason,
332 uint64_t *stub, size_t stub_size,
333 const struct nlattr **actionsp,
334 size_t *actions_lenp);
336 static void xlate_report(struct action_xlate_ctx *ctx, const char *s);
338 /* A subfacet (see "struct subfacet" below) has three possible installation
341 * - SF_NOT_INSTALLED: Not installed in the datapath. This will only be the
342 * case just after the subfacet is created, just before the subfacet is
343 * destroyed, or if the datapath returns an error when we try to install a
346 * - SF_FAST_PATH: The subfacet's actions are installed in the datapath.
348 * - SF_SLOW_PATH: An action that sends every packet for the subfacet through
349 * ofproto_dpif is installed in the datapath.
352 SF_NOT_INSTALLED, /* No datapath flow for this subfacet. */
353 SF_FAST_PATH, /* Full actions are installed. */
354 SF_SLOW_PATH, /* Send-to-userspace action is installed. */
357 static const char *subfacet_path_to_string(enum subfacet_path);
359 /* A dpif flow and actions associated with a facet.
361 * See also the large comment on struct facet. */
364 struct hmap_node hmap_node; /* In struct ofproto_dpif 'subfacets' list. */
365 struct list list_node; /* In struct facet's 'facets' list. */
366 struct facet *facet; /* Owning facet. */
368 enum odp_key_fitness key_fitness;
372 long long int used; /* Time last used; time created if not used. */
374 uint64_t dp_packet_count; /* Last known packet count in the datapath. */
375 uint64_t dp_byte_count; /* Last known byte count in the datapath. */
379 * These should be essentially identical for every subfacet in a facet, but
380 * may differ in trivial ways due to VLAN splinters. */
381 size_t actions_len; /* Number of bytes in actions[]. */
382 struct nlattr *actions; /* Datapath actions. */
384 enum slow_path_reason slow; /* 0 if fast path may be used. */
385 enum subfacet_path path; /* Installed in datapath? */
387 /* Initial values of the packet that may be needed later. */
388 struct initial_vals initial_vals;
390 /* Datapath port the packet arrived on. This is needed to remove
391 * flows for ports that are no longer part of the bridge. Since the
392 * flow definition only has the OpenFlow port number and the port is
393 * no longer part of the bridge, we can't determine the datapath port
394 * number needed to delete the flow from the datapath. */
395 uint32_t odp_in_port;
398 #define SUBFACET_DESTROY_MAX_BATCH 50
400 static struct subfacet *subfacet_create(struct facet *, struct flow_miss *miss,
402 static struct subfacet *subfacet_find(struct ofproto_dpif *,
403 const struct nlattr *key, size_t key_len,
405 static void subfacet_destroy(struct subfacet *);
406 static void subfacet_destroy__(struct subfacet *);
407 static void subfacet_destroy_batch(struct ofproto_dpif *,
408 struct subfacet **, int n);
409 static void subfacet_reset_dp_stats(struct subfacet *,
410 struct dpif_flow_stats *);
411 static void subfacet_update_time(struct subfacet *, long long int used);
412 static void subfacet_update_stats(struct subfacet *,
413 const struct dpif_flow_stats *);
414 static void subfacet_make_actions(struct subfacet *,
415 const struct ofpbuf *packet,
416 struct ofpbuf *odp_actions);
417 static int subfacet_install(struct subfacet *,
418 const struct nlattr *actions, size_t actions_len,
419 struct dpif_flow_stats *, enum slow_path_reason);
420 static void subfacet_uninstall(struct subfacet *);
422 static enum subfacet_path subfacet_want_path(enum slow_path_reason);
424 /* An exact-match instantiation of an OpenFlow flow.
426 * A facet associates a "struct flow", which represents the Open vSwitch
427 * userspace idea of an exact-match flow, with one or more subfacets. Each
428 * subfacet tracks the datapath's idea of the exact-match flow equivalent to
429 * the facet. When the kernel module (or other dpif implementation) and Open
430 * vSwitch userspace agree on the definition of a flow key, there is exactly
431 * one subfacet per facet. If the dpif implementation supports more-specific
432 * flow matching than userspace, however, a facet can have more than one
433 * subfacet, each of which corresponds to some distinction in flow that
434 * userspace simply doesn't understand.
436 * Flow expiration works in terms of subfacets, so a facet must have at least
437 * one subfacet or it will never expire, leaking memory. */
440 struct hmap_node hmap_node; /* In owning ofproto's 'facets' hmap. */
441 struct list list_node; /* In owning rule's 'facets' list. */
442 struct rule_dpif *rule; /* Owning rule. */
445 struct list subfacets;
446 long long int used; /* Time last used; time created if not used. */
453 * - Do include packets and bytes sent "by hand", e.g. with
456 * - Do include packets and bytes that were obtained from the datapath
457 * when a subfacet's statistics were reset (e.g. dpif_flow_put() with
458 * DPIF_FP_ZERO_STATS).
460 * - Do not include packets or bytes that can be obtained from the
461 * datapath for any existing subfacet.
463 uint64_t packet_count; /* Number of packets received. */
464 uint64_t byte_count; /* Number of bytes received. */
466 /* Resubmit statistics. */
467 uint64_t prev_packet_count; /* Number of packets from last stats push. */
468 uint64_t prev_byte_count; /* Number of bytes from last stats push. */
469 long long int prev_used; /* Used time from last stats push. */
472 uint64_t accounted_bytes; /* Bytes processed by facet_account(). */
473 struct netflow_flow nf_flow; /* Per-flow NetFlow tracking data. */
474 uint8_t tcp_flags; /* TCP flags seen for this 'rule'. */
476 /* Properties of datapath actions.
478 * Every subfacet has its own actions because actions can differ slightly
479 * between splintered and non-splintered subfacets due to the VLAN tag
480 * being initially different (present vs. absent). All of them have these
481 * properties in common so we just store one copy of them here. */
482 bool has_learn; /* Actions include NXAST_LEARN? */
483 bool has_normal; /* Actions output to OFPP_NORMAL? */
484 bool has_fin_timeout; /* Actions include NXAST_FIN_TIMEOUT? */
485 tag_type tags; /* Tags that would require revalidation. */
486 mirror_mask_t mirrors; /* Bitmap of dependent mirrors. */
488 /* Storage for a single subfacet, to reduce malloc() time and space
489 * overhead. (A facet always has at least one subfacet and in the common
490 * case has exactly one subfacet.) */
491 struct subfacet one_subfacet;
494 static struct facet *facet_create(struct rule_dpif *,
495 const struct flow *, uint32_t hash);
496 static void facet_remove(struct facet *);
497 static void facet_free(struct facet *);
499 static struct facet *facet_find(struct ofproto_dpif *,
500 const struct flow *, uint32_t hash);
501 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
502 const struct flow *, uint32_t hash);
503 static void facet_revalidate(struct facet *);
504 static bool facet_check_consistency(struct facet *);
506 static void facet_flush_stats(struct facet *);
508 static void facet_update_time(struct facet *, long long int used);
509 static void facet_reset_counters(struct facet *);
510 static void facet_push_stats(struct facet *);
511 static void facet_learn(struct facet *);
512 static void facet_account(struct facet *);
514 static bool facet_is_controller_flow(struct facet *);
517 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
521 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
522 struct list bundle_node; /* In struct ofbundle's "ports" list. */
523 struct cfm *cfm; /* Connectivity Fault Management, if any. */
524 tag_type tag; /* Tag associated with this port. */
525 bool may_enable; /* May be enabled in bonds. */
526 long long int carrier_seq; /* Carrier status changes. */
527 struct tnl_port *tnl_port; /* Tunnel handle, or null. */
530 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
531 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
532 long long int stp_state_entered;
534 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
536 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
538 * This is deprecated. It is only for compatibility with broken device
539 * drivers in old versions of Linux that do not properly support VLANs when
540 * VLAN devices are not used. When broken device drivers are no longer in
541 * widespread use, we will delete these interfaces. */
542 uint16_t realdev_ofp_port;
546 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
547 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
548 * traffic egressing the 'ofport' with that priority should be marked with. */
549 struct priority_to_dscp {
550 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
551 uint32_t priority; /* Priority of this queue (see struct flow). */
553 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
556 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
558 * This is deprecated. It is only for compatibility with broken device drivers
559 * in old versions of Linux that do not properly support VLANs when VLAN
560 * devices are not used. When broken device drivers are no longer in
561 * widespread use, we will delete these interfaces. */
562 struct vlan_splinter {
563 struct hmap_node realdev_vid_node;
564 struct hmap_node vlandev_node;
565 uint16_t realdev_ofp_port;
566 uint16_t vlandev_ofp_port;
570 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
571 uint32_t realdev, ovs_be16 vlan_tci);
572 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
573 static void vsp_remove(struct ofport_dpif *);
574 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
576 static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *,
578 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
581 static struct ofport_dpif *
582 ofport_dpif_cast(const struct ofport *ofport)
584 ovs_assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
585 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
588 static void port_run(struct ofport_dpif *);
589 static void port_run_fast(struct ofport_dpif *);
590 static void port_wait(struct ofport_dpif *);
591 static int set_cfm(struct ofport *, const struct cfm_settings *);
592 static void ofport_clear_priorities(struct ofport_dpif *);
594 struct dpif_completion {
595 struct list list_node;
596 struct ofoperation *op;
599 /* Extra information about a classifier table.
600 * Currently used just for optimized flow revalidation. */
602 /* If either of these is nonnull, then this table has a form that allows
603 * flows to be tagged to avoid revalidating most flows for the most common
604 * kinds of flow table changes. */
605 struct cls_table *catchall_table; /* Table that wildcards all fields. */
606 struct cls_table *other_table; /* Table with any other wildcard set. */
607 uint32_t basis; /* Keeps each table's tags separate. */
610 /* Reasons that we might need to revalidate every facet, and corresponding
613 * A value of 0 means that there is no need to revalidate.
615 * It would be nice to have some cleaner way to integrate with coverage
616 * counters, but with only a few reasons I guess this is good enough for
618 enum revalidate_reason {
619 REV_RECONFIGURE = 1, /* Switch configuration changed. */
620 REV_STP, /* Spanning tree protocol port status change. */
621 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
622 REV_FLOW_TABLE, /* Flow table changed. */
623 REV_INCONSISTENCY /* Facet self-check failed. */
625 COVERAGE_DEFINE(rev_reconfigure);
626 COVERAGE_DEFINE(rev_stp);
627 COVERAGE_DEFINE(rev_port_toggled);
628 COVERAGE_DEFINE(rev_flow_table);
629 COVERAGE_DEFINE(rev_inconsistency);
631 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
632 * These are datapath flows which have no associated ofproto, if they did we
633 * would use facets. */
635 struct hmap_node hmap_node;
640 /* All datapaths of a given type share a single dpif backer instance. */
645 struct timer next_expiration;
646 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
648 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
650 /* Facet revalidation flags applying to facets which use this backer. */
651 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
652 struct tag_set revalidate_set; /* Revalidate only matching facets. */
654 struct hmap drop_keys; /* Set of dropped odp keys. */
657 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
658 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
660 static void drop_key_clear(struct dpif_backer *);
661 static struct ofport_dpif *
662 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
664 struct ofproto_dpif {
665 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
667 struct dpif_backer *backer;
669 /* Special OpenFlow rules. */
670 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
671 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
677 struct netflow *netflow;
678 struct dpif_sflow *sflow;
679 struct hmap bundles; /* Contains "struct ofbundle"s. */
680 struct mac_learning *ml;
681 struct ofmirror *mirrors[MAX_MIRRORS];
683 bool has_bonded_bundles;
687 struct hmap subfacets;
688 struct governor *governor;
691 struct table_dpif tables[N_TABLES];
693 /* Support for debugging async flow mods. */
694 struct list completions;
696 bool has_bundle_action; /* True when the first bundle action appears. */
697 struct netdev_stats stats; /* To account packets generated and consumed in
702 long long int stp_last_tick;
704 /* VLAN splinters. */
705 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
706 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
709 struct sset ports; /* Set of standard port names. */
710 struct sset ghost_ports; /* Ports with no datapath port. */
711 struct sset port_poll_set; /* Queued names for port_poll() reply. */
712 int port_poll_errno; /* Last errno for port_poll() reply. */
715 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
716 * for debugging the asynchronous flow_mod implementation.) */
719 /* All existing ofproto_dpif instances, indexed by ->up.name. */
720 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
722 static void ofproto_dpif_unixctl_init(void);
724 static struct ofproto_dpif *
725 ofproto_dpif_cast(const struct ofproto *ofproto)
727 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
728 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
731 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
733 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
735 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
736 const struct ofpbuf *,
737 const struct initial_vals *, struct ds *);
739 /* Packet processing. */
740 static void update_learning_table(struct ofproto_dpif *,
741 const struct flow *, int vlan,
744 #define FLOW_MISS_MAX_BATCH 50
745 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
747 /* Flow expiration. */
748 static int expire(struct dpif_backer *);
751 static void send_netflow_active_timeouts(struct ofproto_dpif *);
754 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
755 static size_t compose_sflow_action(const struct ofproto_dpif *,
756 struct ofpbuf *odp_actions,
757 const struct flow *, uint32_t odp_port);
758 static void add_mirror_actions(struct action_xlate_ctx *ctx,
759 const struct flow *flow);
760 /* Global variables. */
761 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
763 /* Initial mappings of port to bridge mappings. */
764 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
766 /* Factory functions. */
769 init(const struct shash *iface_hints)
771 struct shash_node *node;
773 /* Make a local copy, since we don't own 'iface_hints' elements. */
774 SHASH_FOR_EACH(node, iface_hints) {
775 const struct iface_hint *orig_hint = node->data;
776 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
778 new_hint->br_name = xstrdup(orig_hint->br_name);
779 new_hint->br_type = xstrdup(orig_hint->br_type);
780 new_hint->ofp_port = orig_hint->ofp_port;
782 shash_add(&init_ofp_ports, node->name, new_hint);
787 enumerate_types(struct sset *types)
789 dp_enumerate_types(types);
793 enumerate_names(const char *type, struct sset *names)
795 struct ofproto_dpif *ofproto;
798 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
799 if (strcmp(type, ofproto->up.type)) {
802 sset_add(names, ofproto->up.name);
809 del(const char *type, const char *name)
814 error = dpif_open(name, type, &dpif);
816 error = dpif_delete(dpif);
823 port_open_type(const char *datapath_type, const char *port_type)
825 return dpif_port_open_type(datapath_type, port_type);
828 /* Type functions. */
830 static struct ofproto_dpif *
831 lookup_ofproto_dpif_by_port_name(const char *name)
833 struct ofproto_dpif *ofproto;
835 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
836 if (sset_contains(&ofproto->ports, name)) {
845 type_run(const char *type)
847 struct dpif_backer *backer;
851 backer = shash_find_data(&all_dpif_backers, type);
853 /* This is not necessarily a problem, since backers are only
854 * created on demand. */
858 dpif_run(backer->dpif);
860 if (backer->need_revalidate
861 || !tag_set_is_empty(&backer->revalidate_set)) {
862 struct tag_set revalidate_set = backer->revalidate_set;
863 bool need_revalidate = backer->need_revalidate;
864 struct ofproto_dpif *ofproto;
865 struct simap_node *node;
866 struct simap tmp_backers;
868 /* Handle tunnel garbage collection. */
869 simap_init(&tmp_backers);
870 simap_swap(&backer->tnl_backers, &tmp_backers);
872 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
873 struct ofport_dpif *iter;
875 if (backer != ofproto->backer) {
879 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
882 if (!iter->tnl_port) {
886 dp_port = netdev_vport_get_dpif_port(iter->up.netdev);
887 node = simap_find(&tmp_backers, dp_port);
889 simap_put(&backer->tnl_backers, dp_port, node->data);
890 simap_delete(&tmp_backers, node);
891 node = simap_find(&backer->tnl_backers, dp_port);
893 node = simap_find(&backer->tnl_backers, dp_port);
895 uint32_t odp_port = UINT32_MAX;
897 if (!dpif_port_add(backer->dpif, iter->up.netdev,
899 simap_put(&backer->tnl_backers, dp_port, odp_port);
900 node = simap_find(&backer->tnl_backers, dp_port);
905 iter->odp_port = node ? node->data : OVSP_NONE;
906 if (tnl_port_reconfigure(&iter->up, iter->odp_port,
908 backer->need_revalidate = REV_RECONFIGURE;
913 SIMAP_FOR_EACH (node, &tmp_backers) {
914 dpif_port_del(backer->dpif, node->data);
916 simap_destroy(&tmp_backers);
918 switch (backer->need_revalidate) {
919 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
920 case REV_STP: COVERAGE_INC(rev_stp); break;
921 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
922 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
923 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
926 if (backer->need_revalidate) {
927 /* Clear the drop_keys in case we should now be accepting some
928 * formerly dropped flows. */
929 drop_key_clear(backer);
932 /* Clear the revalidation flags. */
933 tag_set_init(&backer->revalidate_set);
934 backer->need_revalidate = 0;
936 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
937 struct facet *facet, *next;
939 if (ofproto->backer != backer) {
943 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
945 || tag_set_intersects(&revalidate_set, facet->tags)) {
946 facet_revalidate(facet);
952 if (timer_expired(&backer->next_expiration)) {
953 int delay = expire(backer);
954 timer_set_duration(&backer->next_expiration, delay);
957 /* Check for port changes in the dpif. */
958 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
959 struct ofproto_dpif *ofproto;
960 struct dpif_port port;
962 /* Don't report on the datapath's device. */
963 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
967 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
968 &all_ofproto_dpifs) {
969 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
974 ofproto = lookup_ofproto_dpif_by_port_name(devname);
975 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
976 /* The port was removed. If we know the datapath,
977 * report it through poll_set(). If we don't, it may be
978 * notifying us of a removal we initiated, so ignore it.
979 * If there's a pending ENOBUFS, let it stand, since
980 * everything will be reevaluated. */
981 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
982 sset_add(&ofproto->port_poll_set, devname);
983 ofproto->port_poll_errno = 0;
985 } else if (!ofproto) {
986 /* The port was added, but we don't know with which
987 * ofproto we should associate it. Delete it. */
988 dpif_port_del(backer->dpif, port.port_no);
990 dpif_port_destroy(&port);
996 if (error != EAGAIN) {
997 struct ofproto_dpif *ofproto;
999 /* There was some sort of error, so propagate it to all
1000 * ofprotos that use this backer. */
1001 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1002 &all_ofproto_dpifs) {
1003 if (ofproto->backer == backer) {
1004 sset_clear(&ofproto->port_poll_set);
1005 ofproto->port_poll_errno = error;
1014 type_run_fast(const char *type)
1016 struct dpif_backer *backer;
1019 backer = shash_find_data(&all_dpif_backers, type);
1021 /* This is not necessarily a problem, since backers are only
1022 * created on demand. */
1026 /* Handle one or more batches of upcalls, until there's nothing left to do
1027 * or until we do a fixed total amount of work.
1029 * We do work in batches because it can be much cheaper to set up a number
1030 * of flows and fire off their patches all at once. We do multiple batches
1031 * because in some cases handling a packet can cause another packet to be
1032 * queued almost immediately as part of the return flow. Both
1033 * optimizations can make major improvements on some benchmarks and
1034 * presumably for real traffic as well. */
1036 while (work < FLOW_MISS_MAX_BATCH) {
1037 int retval = handle_upcalls(backer, FLOW_MISS_MAX_BATCH - work);
1048 type_wait(const char *type)
1050 struct dpif_backer *backer;
1052 backer = shash_find_data(&all_dpif_backers, type);
1054 /* This is not necessarily a problem, since backers are only
1055 * created on demand. */
1059 timer_wait(&backer->next_expiration);
1062 /* Basic life-cycle. */
1064 static int add_internal_flows(struct ofproto_dpif *);
1066 static struct ofproto *
1069 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1070 return &ofproto->up;
1074 dealloc(struct ofproto *ofproto_)
1076 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1081 close_dpif_backer(struct dpif_backer *backer)
1083 struct shash_node *node;
1085 ovs_assert(backer->refcount > 0);
1087 if (--backer->refcount) {
1091 drop_key_clear(backer);
1092 hmap_destroy(&backer->drop_keys);
1094 simap_destroy(&backer->tnl_backers);
1095 hmap_destroy(&backer->odp_to_ofport_map);
1096 node = shash_find(&all_dpif_backers, backer->type);
1098 shash_delete(&all_dpif_backers, node);
1099 dpif_close(backer->dpif);
1104 /* Datapath port slated for removal from datapath. */
1105 struct odp_garbage {
1106 struct list list_node;
1111 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1113 struct dpif_backer *backer;
1114 struct dpif_port_dump port_dump;
1115 struct dpif_port port;
1116 struct shash_node *node;
1117 struct list garbage_list;
1118 struct odp_garbage *garbage, *next;
1124 backer = shash_find_data(&all_dpif_backers, type);
1131 backer_name = xasprintf("ovs-%s", type);
1133 /* Remove any existing datapaths, since we assume we're the only
1134 * userspace controlling the datapath. */
1136 dp_enumerate_names(type, &names);
1137 SSET_FOR_EACH(name, &names) {
1138 struct dpif *old_dpif;
1140 /* Don't remove our backer if it exists. */
1141 if (!strcmp(name, backer_name)) {
1145 if (dpif_open(name, type, &old_dpif)) {
1146 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1148 dpif_delete(old_dpif);
1149 dpif_close(old_dpif);
1152 sset_destroy(&names);
1154 backer = xmalloc(sizeof *backer);
1156 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1159 VLOG_ERR("failed to open datapath of type %s: %s", type,
1165 backer->type = xstrdup(type);
1166 backer->refcount = 1;
1167 hmap_init(&backer->odp_to_ofport_map);
1168 hmap_init(&backer->drop_keys);
1169 timer_set_duration(&backer->next_expiration, 1000);
1170 backer->need_revalidate = 0;
1171 simap_init(&backer->tnl_backers);
1172 tag_set_init(&backer->revalidate_set);
1175 dpif_flow_flush(backer->dpif);
1177 /* Loop through the ports already on the datapath and remove any
1178 * that we don't need anymore. */
1179 list_init(&garbage_list);
1180 dpif_port_dump_start(&port_dump, backer->dpif);
1181 while (dpif_port_dump_next(&port_dump, &port)) {
1182 node = shash_find(&init_ofp_ports, port.name);
1183 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1184 garbage = xmalloc(sizeof *garbage);
1185 garbage->odp_port = port.port_no;
1186 list_push_front(&garbage_list, &garbage->list_node);
1189 dpif_port_dump_done(&port_dump);
1191 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1192 dpif_port_del(backer->dpif, garbage->odp_port);
1193 list_remove(&garbage->list_node);
1197 shash_add(&all_dpif_backers, type, backer);
1199 error = dpif_recv_set(backer->dpif, true);
1201 VLOG_ERR("failed to listen on datapath of type %s: %s",
1202 type, strerror(error));
1203 close_dpif_backer(backer);
1211 construct(struct ofproto *ofproto_)
1213 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1214 struct shash_node *node, *next;
1219 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1224 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1225 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1227 ofproto->n_matches = 0;
1229 ofproto->netflow = NULL;
1230 ofproto->sflow = NULL;
1231 ofproto->stp = NULL;
1232 hmap_init(&ofproto->bundles);
1233 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1234 for (i = 0; i < MAX_MIRRORS; i++) {
1235 ofproto->mirrors[i] = NULL;
1237 ofproto->has_bonded_bundles = false;
1239 hmap_init(&ofproto->facets);
1240 hmap_init(&ofproto->subfacets);
1241 ofproto->governor = NULL;
1243 for (i = 0; i < N_TABLES; i++) {
1244 struct table_dpif *table = &ofproto->tables[i];
1246 table->catchall_table = NULL;
1247 table->other_table = NULL;
1248 table->basis = random_uint32();
1251 list_init(&ofproto->completions);
1253 ofproto_dpif_unixctl_init();
1255 ofproto->has_mirrors = false;
1256 ofproto->has_bundle_action = false;
1258 hmap_init(&ofproto->vlandev_map);
1259 hmap_init(&ofproto->realdev_vid_map);
1261 sset_init(&ofproto->ports);
1262 sset_init(&ofproto->ghost_ports);
1263 sset_init(&ofproto->port_poll_set);
1264 ofproto->port_poll_errno = 0;
1266 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1267 struct iface_hint *iface_hint = node->data;
1269 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1270 /* Check if the datapath already has this port. */
1271 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1272 sset_add(&ofproto->ports, node->name);
1275 free(iface_hint->br_name);
1276 free(iface_hint->br_type);
1278 shash_delete(&init_ofp_ports, node);
1282 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1283 hash_string(ofproto->up.name, 0));
1284 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1286 ofproto_init_tables(ofproto_, N_TABLES);
1287 error = add_internal_flows(ofproto);
1288 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1294 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1295 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1297 struct ofputil_flow_mod fm;
1300 match_init_catchall(&fm.match);
1302 match_set_reg(&fm.match, 0, id);
1303 fm.new_cookie = htonll(0);
1304 fm.cookie = htonll(0);
1305 fm.cookie_mask = htonll(0);
1306 fm.table_id = TBL_INTERNAL;
1307 fm.command = OFPFC_ADD;
1308 fm.idle_timeout = 0;
1309 fm.hard_timeout = 0;
1313 fm.ofpacts = ofpacts->data;
1314 fm.ofpacts_len = ofpacts->size;
1316 error = ofproto_flow_mod(&ofproto->up, &fm);
1318 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1319 id, ofperr_to_string(error));
1323 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1324 ovs_assert(*rulep != NULL);
1330 add_internal_flows(struct ofproto_dpif *ofproto)
1332 struct ofpact_controller *controller;
1333 uint64_t ofpacts_stub[128 / 8];
1334 struct ofpbuf ofpacts;
1338 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1341 controller = ofpact_put_CONTROLLER(&ofpacts);
1342 controller->max_len = UINT16_MAX;
1343 controller->controller_id = 0;
1344 controller->reason = OFPR_NO_MATCH;
1345 ofpact_pad(&ofpacts);
1347 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1352 ofpbuf_clear(&ofpacts);
1353 error = add_internal_flow(ofproto, id++, &ofpacts,
1354 &ofproto->no_packet_in_rule);
1359 complete_operations(struct ofproto_dpif *ofproto)
1361 struct dpif_completion *c, *next;
1363 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1364 ofoperation_complete(c->op, 0);
1365 list_remove(&c->list_node);
1371 destruct(struct ofproto *ofproto_)
1373 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1374 struct rule_dpif *rule, *next_rule;
1375 struct oftable *table;
1378 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1379 complete_operations(ofproto);
1381 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1382 struct cls_cursor cursor;
1384 cls_cursor_init(&cursor, &table->cls, NULL);
1385 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1386 ofproto_rule_destroy(&rule->up);
1390 for (i = 0; i < MAX_MIRRORS; i++) {
1391 mirror_destroy(ofproto->mirrors[i]);
1394 netflow_destroy(ofproto->netflow);
1395 dpif_sflow_destroy(ofproto->sflow);
1396 hmap_destroy(&ofproto->bundles);
1397 mac_learning_destroy(ofproto->ml);
1399 hmap_destroy(&ofproto->facets);
1400 hmap_destroy(&ofproto->subfacets);
1401 governor_destroy(ofproto->governor);
1403 hmap_destroy(&ofproto->vlandev_map);
1404 hmap_destroy(&ofproto->realdev_vid_map);
1406 sset_destroy(&ofproto->ports);
1407 sset_destroy(&ofproto->ghost_ports);
1408 sset_destroy(&ofproto->port_poll_set);
1410 close_dpif_backer(ofproto->backer);
1414 run_fast(struct ofproto *ofproto_)
1416 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1417 struct ofport_dpif *ofport;
1419 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1420 port_run_fast(ofport);
1427 run(struct ofproto *ofproto_)
1429 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1430 struct ofport_dpif *ofport;
1431 struct ofbundle *bundle;
1435 complete_operations(ofproto);
1438 error = run_fast(ofproto_);
1443 if (ofproto->netflow) {
1444 if (netflow_run(ofproto->netflow)) {
1445 send_netflow_active_timeouts(ofproto);
1448 if (ofproto->sflow) {
1449 dpif_sflow_run(ofproto->sflow);
1452 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1455 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1460 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1462 /* Check the consistency of a random facet, to aid debugging. */
1463 if (!hmap_is_empty(&ofproto->facets)
1464 && !ofproto->backer->need_revalidate) {
1465 struct facet *facet;
1467 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1468 struct facet, hmap_node);
1469 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1471 if (!facet_check_consistency(facet)) {
1472 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1477 if (ofproto->governor) {
1480 governor_run(ofproto->governor);
1482 /* If the governor has shrunk to its minimum size and the number of
1483 * subfacets has dwindled, then drop the governor entirely.
1485 * For hysteresis, the number of subfacets to drop the governor is
1486 * smaller than the number needed to trigger its creation. */
1487 n_subfacets = hmap_count(&ofproto->subfacets);
1488 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1489 && governor_is_idle(ofproto->governor)) {
1490 governor_destroy(ofproto->governor);
1491 ofproto->governor = NULL;
1499 wait(struct ofproto *ofproto_)
1501 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1502 struct ofport_dpif *ofport;
1503 struct ofbundle *bundle;
1505 if (!clogged && !list_is_empty(&ofproto->completions)) {
1506 poll_immediate_wake();
1509 dpif_wait(ofproto->backer->dpif);
1510 dpif_recv_wait(ofproto->backer->dpif);
1511 if (ofproto->sflow) {
1512 dpif_sflow_wait(ofproto->sflow);
1514 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1515 poll_immediate_wake();
1517 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1520 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1521 bundle_wait(bundle);
1523 if (ofproto->netflow) {
1524 netflow_wait(ofproto->netflow);
1526 mac_learning_wait(ofproto->ml);
1528 if (ofproto->backer->need_revalidate) {
1529 /* Shouldn't happen, but if it does just go around again. */
1530 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1531 poll_immediate_wake();
1533 if (ofproto->governor) {
1534 governor_wait(ofproto->governor);
1539 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1541 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1543 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1544 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1548 flush(struct ofproto *ofproto_)
1550 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1551 struct subfacet *subfacet, *next_subfacet;
1552 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1556 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1557 &ofproto->subfacets) {
1558 if (subfacet->path != SF_NOT_INSTALLED) {
1559 batch[n_batch++] = subfacet;
1560 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1561 subfacet_destroy_batch(ofproto, batch, n_batch);
1565 subfacet_destroy(subfacet);
1570 subfacet_destroy_batch(ofproto, batch, n_batch);
1575 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1576 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1578 *arp_match_ip = true;
1579 *actions = (OFPUTIL_A_OUTPUT |
1580 OFPUTIL_A_SET_VLAN_VID |
1581 OFPUTIL_A_SET_VLAN_PCP |
1582 OFPUTIL_A_STRIP_VLAN |
1583 OFPUTIL_A_SET_DL_SRC |
1584 OFPUTIL_A_SET_DL_DST |
1585 OFPUTIL_A_SET_NW_SRC |
1586 OFPUTIL_A_SET_NW_DST |
1587 OFPUTIL_A_SET_NW_TOS |
1588 OFPUTIL_A_SET_TP_SRC |
1589 OFPUTIL_A_SET_TP_DST |
1594 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1596 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1597 struct dpif_dp_stats s;
1599 strcpy(ots->name, "classifier");
1601 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1603 ots->lookup_count = htonll(s.n_hit + s.n_missed);
1604 ots->matched_count = htonll(s.n_hit + ofproto->n_matches);
1607 static struct ofport *
1610 struct ofport_dpif *port = xmalloc(sizeof *port);
1615 port_dealloc(struct ofport *port_)
1617 struct ofport_dpif *port = ofport_dpif_cast(port_);
1622 port_construct(struct ofport *port_)
1624 struct ofport_dpif *port = ofport_dpif_cast(port_);
1625 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1626 const struct netdev *netdev = port->up.netdev;
1627 struct dpif_port dpif_port;
1630 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1631 port->bundle = NULL;
1633 port->tag = tag_create_random();
1634 port->may_enable = true;
1635 port->stp_port = NULL;
1636 port->stp_state = STP_DISABLED;
1637 port->tnl_port = NULL;
1638 hmap_init(&port->priorities);
1639 port->realdev_ofp_port = 0;
1640 port->vlandev_vid = 0;
1641 port->carrier_seq = netdev_get_carrier_resets(netdev);
1643 if (netdev_vport_is_patch(netdev)) {
1644 /* XXX By bailing out here, we don't do required sFlow work. */
1645 port->odp_port = OVSP_NONE;
1649 error = dpif_port_query_by_name(ofproto->backer->dpif,
1650 netdev_vport_get_dpif_port(netdev),
1656 port->odp_port = dpif_port.port_no;
1658 if (netdev_get_tunnel_config(netdev)) {
1659 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1661 /* Sanity-check that a mapping doesn't already exist. This
1662 * shouldn't happen for non-tunnel ports. */
1663 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1664 VLOG_ERR("port %s already has an OpenFlow port number",
1666 dpif_port_destroy(&dpif_port);
1670 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1671 hash_int(port->odp_port, 0));
1673 dpif_port_destroy(&dpif_port);
1675 if (ofproto->sflow) {
1676 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1683 port_destruct(struct ofport *port_)
1685 struct ofport_dpif *port = ofport_dpif_cast(port_);
1686 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1687 const char *dp_port_name = netdev_vport_get_dpif_port(port->up.netdev);
1688 const char *devname = netdev_get_name(port->up.netdev);
1690 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1691 /* The underlying device is still there, so delete it. This
1692 * happens when the ofproto is being destroyed, since the caller
1693 * assumes that removal of attached ports will happen as part of
1695 if (!port->tnl_port) {
1696 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1698 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1701 if (port->odp_port != OVSP_NONE && !port->tnl_port) {
1702 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1705 tnl_port_del(port->tnl_port);
1706 sset_find_and_delete(&ofproto->ports, devname);
1707 sset_find_and_delete(&ofproto->ghost_ports, devname);
1708 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1709 bundle_remove(port_);
1710 set_cfm(port_, NULL);
1711 if (ofproto->sflow) {
1712 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1715 ofport_clear_priorities(port);
1716 hmap_destroy(&port->priorities);
1720 port_modified(struct ofport *port_)
1722 struct ofport_dpif *port = ofport_dpif_cast(port_);
1724 if (port->bundle && port->bundle->bond) {
1725 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1730 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1732 struct ofport_dpif *port = ofport_dpif_cast(port_);
1733 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1734 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1736 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1737 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1738 OFPUTIL_PC_NO_PACKET_IN)) {
1739 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1741 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1742 bundle_update(port->bundle);
1748 set_sflow(struct ofproto *ofproto_,
1749 const struct ofproto_sflow_options *sflow_options)
1751 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1752 struct dpif_sflow *ds = ofproto->sflow;
1754 if (sflow_options) {
1756 struct ofport_dpif *ofport;
1758 ds = ofproto->sflow = dpif_sflow_create();
1759 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1760 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1762 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1764 dpif_sflow_set_options(ds, sflow_options);
1767 dpif_sflow_destroy(ds);
1768 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1769 ofproto->sflow = NULL;
1776 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1778 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1785 struct ofproto_dpif *ofproto;
1787 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1788 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1789 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1792 if (cfm_configure(ofport->cfm, s)) {
1798 cfm_destroy(ofport->cfm);
1804 get_cfm_status(const struct ofport *ofport_,
1805 struct ofproto_cfm_status *status)
1807 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1810 status->faults = cfm_get_fault(ofport->cfm);
1811 status->remote_opstate = cfm_get_opup(ofport->cfm);
1812 status->health = cfm_get_health(ofport->cfm);
1813 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1820 /* Spanning Tree. */
1823 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1825 struct ofproto_dpif *ofproto = ofproto_;
1826 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1827 struct ofport_dpif *ofport;
1829 ofport = stp_port_get_aux(sp);
1831 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1832 ofproto->up.name, port_num);
1834 struct eth_header *eth = pkt->l2;
1836 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1837 if (eth_addr_is_zero(eth->eth_src)) {
1838 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1839 "with unknown MAC", ofproto->up.name, port_num);
1841 send_packet(ofport, pkt);
1847 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1849 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1851 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1853 /* Only revalidate flows if the configuration changed. */
1854 if (!s != !ofproto->stp) {
1855 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1859 if (!ofproto->stp) {
1860 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1861 send_bpdu_cb, ofproto);
1862 ofproto->stp_last_tick = time_msec();
1865 stp_set_bridge_id(ofproto->stp, s->system_id);
1866 stp_set_bridge_priority(ofproto->stp, s->priority);
1867 stp_set_hello_time(ofproto->stp, s->hello_time);
1868 stp_set_max_age(ofproto->stp, s->max_age);
1869 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1871 struct ofport *ofport;
1873 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1874 set_stp_port(ofport, NULL);
1877 stp_destroy(ofproto->stp);
1878 ofproto->stp = NULL;
1885 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1887 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1891 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1892 s->designated_root = stp_get_designated_root(ofproto->stp);
1893 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1902 update_stp_port_state(struct ofport_dpif *ofport)
1904 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1905 enum stp_state state;
1907 /* Figure out new state. */
1908 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1912 if (ofport->stp_state != state) {
1913 enum ofputil_port_state of_state;
1916 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1917 netdev_get_name(ofport->up.netdev),
1918 stp_state_name(ofport->stp_state),
1919 stp_state_name(state));
1920 if (stp_learn_in_state(ofport->stp_state)
1921 != stp_learn_in_state(state)) {
1922 /* xxx Learning action flows should also be flushed. */
1923 mac_learning_flush(ofproto->ml,
1924 &ofproto->backer->revalidate_set);
1926 fwd_change = stp_forward_in_state(ofport->stp_state)
1927 != stp_forward_in_state(state);
1929 ofproto->backer->need_revalidate = REV_STP;
1930 ofport->stp_state = state;
1931 ofport->stp_state_entered = time_msec();
1933 if (fwd_change && ofport->bundle) {
1934 bundle_update(ofport->bundle);
1937 /* Update the STP state bits in the OpenFlow port description. */
1938 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1939 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1940 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1941 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1942 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1944 ofproto_port_set_state(&ofport->up, of_state);
1948 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1949 * caller is responsible for assigning STP port numbers and ensuring
1950 * there are no duplicates. */
1952 set_stp_port(struct ofport *ofport_,
1953 const struct ofproto_port_stp_settings *s)
1955 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1956 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1957 struct stp_port *sp = ofport->stp_port;
1959 if (!s || !s->enable) {
1961 ofport->stp_port = NULL;
1962 stp_port_disable(sp);
1963 update_stp_port_state(ofport);
1966 } else if (sp && stp_port_no(sp) != s->port_num
1967 && ofport == stp_port_get_aux(sp)) {
1968 /* The port-id changed, so disable the old one if it's not
1969 * already in use by another port. */
1970 stp_port_disable(sp);
1973 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1974 stp_port_enable(sp);
1976 stp_port_set_aux(sp, ofport);
1977 stp_port_set_priority(sp, s->priority);
1978 stp_port_set_path_cost(sp, s->path_cost);
1980 update_stp_port_state(ofport);
1986 get_stp_port_status(struct ofport *ofport_,
1987 struct ofproto_port_stp_status *s)
1989 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1990 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1991 struct stp_port *sp = ofport->stp_port;
1993 if (!ofproto->stp || !sp) {
1999 s->port_id = stp_port_get_id(sp);
2000 s->state = stp_port_get_state(sp);
2001 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2002 s->role = stp_port_get_role(sp);
2003 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2009 stp_run(struct ofproto_dpif *ofproto)
2012 long long int now = time_msec();
2013 long long int elapsed = now - ofproto->stp_last_tick;
2014 struct stp_port *sp;
2017 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2018 ofproto->stp_last_tick = now;
2020 while (stp_get_changed_port(ofproto->stp, &sp)) {
2021 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2024 update_stp_port_state(ofport);
2028 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2029 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2035 stp_wait(struct ofproto_dpif *ofproto)
2038 poll_timer_wait(1000);
2042 /* Returns true if STP should process 'flow'. */
2044 stp_should_process_flow(const struct flow *flow)
2046 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2050 stp_process_packet(const struct ofport_dpif *ofport,
2051 const struct ofpbuf *packet)
2053 struct ofpbuf payload = *packet;
2054 struct eth_header *eth = payload.data;
2055 struct stp_port *sp = ofport->stp_port;
2057 /* Sink packets on ports that have STP disabled when the bridge has
2059 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2063 /* Trim off padding on payload. */
2064 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2065 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2068 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2069 stp_received_bpdu(sp, payload.data, payload.size);
2073 static struct priority_to_dscp *
2074 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2076 struct priority_to_dscp *pdscp;
2079 hash = hash_int(priority, 0);
2080 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2081 if (pdscp->priority == priority) {
2089 ofport_clear_priorities(struct ofport_dpif *ofport)
2091 struct priority_to_dscp *pdscp, *next;
2093 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2094 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2100 set_queues(struct ofport *ofport_,
2101 const struct ofproto_port_queue *qdscp_list,
2104 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2105 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2106 struct hmap new = HMAP_INITIALIZER(&new);
2109 for (i = 0; i < n_qdscp; i++) {
2110 struct priority_to_dscp *pdscp;
2114 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2115 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2120 pdscp = get_priority(ofport, priority);
2122 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2124 pdscp = xmalloc(sizeof *pdscp);
2125 pdscp->priority = priority;
2127 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2130 if (pdscp->dscp != dscp) {
2132 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2135 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2138 if (!hmap_is_empty(&ofport->priorities)) {
2139 ofport_clear_priorities(ofport);
2140 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2143 hmap_swap(&new, &ofport->priorities);
2151 /* Expires all MAC learning entries associated with 'bundle' and forces its
2152 * ofproto to revalidate every flow.
2154 * Normally MAC learning entries are removed only from the ofproto associated
2155 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2156 * are removed from every ofproto. When patch ports and SLB bonds are in use
2157 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2158 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2159 * with the host from which it migrated. */
2161 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2163 struct ofproto_dpif *ofproto = bundle->ofproto;
2164 struct mac_learning *ml = ofproto->ml;
2165 struct mac_entry *mac, *next_mac;
2167 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2168 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2169 if (mac->port.p == bundle) {
2171 struct ofproto_dpif *o;
2173 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2175 struct mac_entry *e;
2177 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2180 mac_learning_expire(o->ml, e);
2186 mac_learning_expire(ml, mac);
2191 static struct ofbundle *
2192 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2194 struct ofbundle *bundle;
2196 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2197 &ofproto->bundles) {
2198 if (bundle->aux == aux) {
2205 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2206 * ones that are found to 'bundles'. */
2208 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2209 void **auxes, size_t n_auxes,
2210 struct hmapx *bundles)
2214 hmapx_init(bundles);
2215 for (i = 0; i < n_auxes; i++) {
2216 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2218 hmapx_add(bundles, bundle);
2224 bundle_update(struct ofbundle *bundle)
2226 struct ofport_dpif *port;
2228 bundle->floodable = true;
2229 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2230 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2231 || !stp_forward_in_state(port->stp_state)) {
2232 bundle->floodable = false;
2239 bundle_del_port(struct ofport_dpif *port)
2241 struct ofbundle *bundle = port->bundle;
2243 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2245 list_remove(&port->bundle_node);
2246 port->bundle = NULL;
2249 lacp_slave_unregister(bundle->lacp, port);
2252 bond_slave_unregister(bundle->bond, port);
2255 bundle_update(bundle);
2259 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
2260 struct lacp_slave_settings *lacp)
2262 struct ofport_dpif *port;
2264 port = get_ofp_port(bundle->ofproto, ofp_port);
2269 if (port->bundle != bundle) {
2270 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2272 bundle_del_port(port);
2275 port->bundle = bundle;
2276 list_push_back(&bundle->ports, &port->bundle_node);
2277 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2278 || !stp_forward_in_state(port->stp_state)) {
2279 bundle->floodable = false;
2283 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2284 lacp_slave_register(bundle->lacp, port, lacp);
2291 bundle_destroy(struct ofbundle *bundle)
2293 struct ofproto_dpif *ofproto;
2294 struct ofport_dpif *port, *next_port;
2301 ofproto = bundle->ofproto;
2302 for (i = 0; i < MAX_MIRRORS; i++) {
2303 struct ofmirror *m = ofproto->mirrors[i];
2305 if (m->out == bundle) {
2307 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2308 || hmapx_find_and_delete(&m->dsts, bundle)) {
2309 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2314 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2315 bundle_del_port(port);
2318 bundle_flush_macs(bundle, true);
2319 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2321 free(bundle->trunks);
2322 lacp_destroy(bundle->lacp);
2323 bond_destroy(bundle->bond);
2328 bundle_set(struct ofproto *ofproto_, void *aux,
2329 const struct ofproto_bundle_settings *s)
2331 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2332 bool need_flush = false;
2333 struct ofport_dpif *port;
2334 struct ofbundle *bundle;
2335 unsigned long *trunks;
2341 bundle_destroy(bundle_lookup(ofproto, aux));
2345 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2346 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2348 bundle = bundle_lookup(ofproto, aux);
2350 bundle = xmalloc(sizeof *bundle);
2352 bundle->ofproto = ofproto;
2353 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2354 hash_pointer(aux, 0));
2356 bundle->name = NULL;
2358 list_init(&bundle->ports);
2359 bundle->vlan_mode = PORT_VLAN_TRUNK;
2361 bundle->trunks = NULL;
2362 bundle->use_priority_tags = s->use_priority_tags;
2363 bundle->lacp = NULL;
2364 bundle->bond = NULL;
2366 bundle->floodable = true;
2368 bundle->src_mirrors = 0;
2369 bundle->dst_mirrors = 0;
2370 bundle->mirror_out = 0;
2373 if (!bundle->name || strcmp(s->name, bundle->name)) {
2375 bundle->name = xstrdup(s->name);
2380 if (!bundle->lacp) {
2381 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2382 bundle->lacp = lacp_create();
2384 lacp_configure(bundle->lacp, s->lacp);
2386 lacp_destroy(bundle->lacp);
2387 bundle->lacp = NULL;
2390 /* Update set of ports. */
2392 for (i = 0; i < s->n_slaves; i++) {
2393 if (!bundle_add_port(bundle, s->slaves[i],
2394 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2398 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2399 struct ofport_dpif *next_port;
2401 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2402 for (i = 0; i < s->n_slaves; i++) {
2403 if (s->slaves[i] == port->up.ofp_port) {
2408 bundle_del_port(port);
2412 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2414 if (list_is_empty(&bundle->ports)) {
2415 bundle_destroy(bundle);
2419 /* Set VLAN tagging mode */
2420 if (s->vlan_mode != bundle->vlan_mode
2421 || s->use_priority_tags != bundle->use_priority_tags) {
2422 bundle->vlan_mode = s->vlan_mode;
2423 bundle->use_priority_tags = s->use_priority_tags;
2428 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2429 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2431 if (vlan != bundle->vlan) {
2432 bundle->vlan = vlan;
2436 /* Get trunked VLANs. */
2437 switch (s->vlan_mode) {
2438 case PORT_VLAN_ACCESS:
2442 case PORT_VLAN_TRUNK:
2443 trunks = CONST_CAST(unsigned long *, s->trunks);
2446 case PORT_VLAN_NATIVE_UNTAGGED:
2447 case PORT_VLAN_NATIVE_TAGGED:
2448 if (vlan != 0 && (!s->trunks
2449 || !bitmap_is_set(s->trunks, vlan)
2450 || bitmap_is_set(s->trunks, 0))) {
2451 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2453 trunks = bitmap_clone(s->trunks, 4096);
2455 trunks = bitmap_allocate1(4096);
2457 bitmap_set1(trunks, vlan);
2458 bitmap_set0(trunks, 0);
2460 trunks = CONST_CAST(unsigned long *, s->trunks);
2467 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2468 free(bundle->trunks);
2469 if (trunks == s->trunks) {
2470 bundle->trunks = vlan_bitmap_clone(trunks);
2472 bundle->trunks = trunks;
2477 if (trunks != s->trunks) {
2482 if (!list_is_short(&bundle->ports)) {
2483 bundle->ofproto->has_bonded_bundles = true;
2485 if (bond_reconfigure(bundle->bond, s->bond)) {
2486 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2489 bundle->bond = bond_create(s->bond);
2490 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2493 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2494 bond_slave_register(bundle->bond, port, port->up.netdev);
2497 bond_destroy(bundle->bond);
2498 bundle->bond = NULL;
2501 /* If we changed something that would affect MAC learning, un-learn
2502 * everything on this port and force flow revalidation. */
2504 bundle_flush_macs(bundle, false);
2511 bundle_remove(struct ofport *port_)
2513 struct ofport_dpif *port = ofport_dpif_cast(port_);
2514 struct ofbundle *bundle = port->bundle;
2517 bundle_del_port(port);
2518 if (list_is_empty(&bundle->ports)) {
2519 bundle_destroy(bundle);
2520 } else if (list_is_short(&bundle->ports)) {
2521 bond_destroy(bundle->bond);
2522 bundle->bond = NULL;
2528 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2530 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2531 struct ofport_dpif *port = port_;
2532 uint8_t ea[ETH_ADDR_LEN];
2535 error = netdev_get_etheraddr(port->up.netdev, ea);
2537 struct ofpbuf packet;
2540 ofpbuf_init(&packet, 0);
2541 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2543 memcpy(packet_pdu, pdu, pdu_size);
2545 send_packet(port, &packet);
2546 ofpbuf_uninit(&packet);
2548 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2549 "%s (%s)", port->bundle->name,
2550 netdev_get_name(port->up.netdev), strerror(error));
2555 bundle_send_learning_packets(struct ofbundle *bundle)
2557 struct ofproto_dpif *ofproto = bundle->ofproto;
2558 int error, n_packets, n_errors;
2559 struct mac_entry *e;
2561 error = n_packets = n_errors = 0;
2562 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2563 if (e->port.p != bundle) {
2564 struct ofpbuf *learning_packet;
2565 struct ofport_dpif *port;
2569 /* The assignment to "port" is unnecessary but makes "grep"ing for
2570 * struct ofport_dpif more effective. */
2571 learning_packet = bond_compose_learning_packet(bundle->bond,
2575 ret = send_packet(port, learning_packet);
2576 ofpbuf_delete(learning_packet);
2586 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2587 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2588 "packets, last error was: %s",
2589 bundle->name, n_errors, n_packets, strerror(error));
2591 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2592 bundle->name, n_packets);
2597 bundle_run(struct ofbundle *bundle)
2600 lacp_run(bundle->lacp, send_pdu_cb);
2603 struct ofport_dpif *port;
2605 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2606 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2609 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2610 lacp_status(bundle->lacp));
2611 if (bond_should_send_learning_packets(bundle->bond)) {
2612 bundle_send_learning_packets(bundle);
2618 bundle_wait(struct ofbundle *bundle)
2621 lacp_wait(bundle->lacp);
2624 bond_wait(bundle->bond);
2631 mirror_scan(struct ofproto_dpif *ofproto)
2635 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2636 if (!ofproto->mirrors[idx]) {
2643 static struct ofmirror *
2644 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2648 for (i = 0; i < MAX_MIRRORS; i++) {
2649 struct ofmirror *mirror = ofproto->mirrors[i];
2650 if (mirror && mirror->aux == aux) {
2658 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2660 mirror_update_dups(struct ofproto_dpif *ofproto)
2664 for (i = 0; i < MAX_MIRRORS; i++) {
2665 struct ofmirror *m = ofproto->mirrors[i];
2668 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2672 for (i = 0; i < MAX_MIRRORS; i++) {
2673 struct ofmirror *m1 = ofproto->mirrors[i];
2680 for (j = i + 1; j < MAX_MIRRORS; j++) {
2681 struct ofmirror *m2 = ofproto->mirrors[j];
2683 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2684 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2685 m2->dup_mirrors |= m1->dup_mirrors;
2692 mirror_set(struct ofproto *ofproto_, void *aux,
2693 const struct ofproto_mirror_settings *s)
2695 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2696 mirror_mask_t mirror_bit;
2697 struct ofbundle *bundle;
2698 struct ofmirror *mirror;
2699 struct ofbundle *out;
2700 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2701 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2704 mirror = mirror_lookup(ofproto, aux);
2706 mirror_destroy(mirror);
2712 idx = mirror_scan(ofproto);
2714 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2716 ofproto->up.name, MAX_MIRRORS, s->name);
2720 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2721 mirror->ofproto = ofproto;
2724 mirror->out_vlan = -1;
2725 mirror->name = NULL;
2728 if (!mirror->name || strcmp(s->name, mirror->name)) {
2730 mirror->name = xstrdup(s->name);
2733 /* Get the new configuration. */
2734 if (s->out_bundle) {
2735 out = bundle_lookup(ofproto, s->out_bundle);
2737 mirror_destroy(mirror);
2743 out_vlan = s->out_vlan;
2745 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2746 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2748 /* If the configuration has not changed, do nothing. */
2749 if (hmapx_equals(&srcs, &mirror->srcs)
2750 && hmapx_equals(&dsts, &mirror->dsts)
2751 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2752 && mirror->out == out
2753 && mirror->out_vlan == out_vlan)
2755 hmapx_destroy(&srcs);
2756 hmapx_destroy(&dsts);
2760 hmapx_swap(&srcs, &mirror->srcs);
2761 hmapx_destroy(&srcs);
2763 hmapx_swap(&dsts, &mirror->dsts);
2764 hmapx_destroy(&dsts);
2766 free(mirror->vlans);
2767 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2770 mirror->out_vlan = out_vlan;
2772 /* Update bundles. */
2773 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2774 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2775 if (hmapx_contains(&mirror->srcs, bundle)) {
2776 bundle->src_mirrors |= mirror_bit;
2778 bundle->src_mirrors &= ~mirror_bit;
2781 if (hmapx_contains(&mirror->dsts, bundle)) {
2782 bundle->dst_mirrors |= mirror_bit;
2784 bundle->dst_mirrors &= ~mirror_bit;
2787 if (mirror->out == bundle) {
2788 bundle->mirror_out |= mirror_bit;
2790 bundle->mirror_out &= ~mirror_bit;
2794 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2795 ofproto->has_mirrors = true;
2796 mac_learning_flush(ofproto->ml,
2797 &ofproto->backer->revalidate_set);
2798 mirror_update_dups(ofproto);
2804 mirror_destroy(struct ofmirror *mirror)
2806 struct ofproto_dpif *ofproto;
2807 mirror_mask_t mirror_bit;
2808 struct ofbundle *bundle;
2815 ofproto = mirror->ofproto;
2816 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2817 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2819 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2820 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2821 bundle->src_mirrors &= ~mirror_bit;
2822 bundle->dst_mirrors &= ~mirror_bit;
2823 bundle->mirror_out &= ~mirror_bit;
2826 hmapx_destroy(&mirror->srcs);
2827 hmapx_destroy(&mirror->dsts);
2828 free(mirror->vlans);
2830 ofproto->mirrors[mirror->idx] = NULL;
2834 mirror_update_dups(ofproto);
2836 ofproto->has_mirrors = false;
2837 for (i = 0; i < MAX_MIRRORS; i++) {
2838 if (ofproto->mirrors[i]) {
2839 ofproto->has_mirrors = true;
2846 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2847 uint64_t *packets, uint64_t *bytes)
2849 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2850 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2853 *packets = *bytes = UINT64_MAX;
2857 *packets = mirror->packet_count;
2858 *bytes = mirror->byte_count;
2864 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2866 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2867 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2868 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2874 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2876 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2877 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2878 return bundle && bundle->mirror_out != 0;
2882 forward_bpdu_changed(struct ofproto *ofproto_)
2884 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2885 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2889 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2892 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2893 mac_learning_set_idle_time(ofproto->ml, idle_time);
2894 mac_learning_set_max_entries(ofproto->ml, max_entries);
2899 static struct ofport_dpif *
2900 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
2902 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2903 return ofport ? ofport_dpif_cast(ofport) : NULL;
2906 static struct ofport_dpif *
2907 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
2909 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2910 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2914 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2915 struct ofproto_port *ofproto_port,
2916 struct dpif_port *dpif_port)
2918 ofproto_port->name = dpif_port->name;
2919 ofproto_port->type = dpif_port->type;
2920 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2923 static struct ofport_dpif *
2924 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
2926 const struct ofproto_dpif *ofproto;
2929 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
2934 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2935 struct ofport *ofport;
2937 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
2938 if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) {
2939 return ofport_dpif_cast(ofport);
2946 port_run_fast(struct ofport_dpif *ofport)
2948 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2949 struct ofpbuf packet;
2951 ofpbuf_init(&packet, 0);
2952 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2953 send_packet(ofport, &packet);
2954 ofpbuf_uninit(&packet);
2959 port_run(struct ofport_dpif *ofport)
2961 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2962 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2963 bool enable = netdev_get_carrier(ofport->up.netdev);
2965 ofport->carrier_seq = carrier_seq;
2967 port_run_fast(ofport);
2969 if (ofport->tnl_port
2970 && tnl_port_reconfigure(&ofport->up, ofport->odp_port,
2971 &ofport->tnl_port)) {
2972 ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true;
2976 int cfm_opup = cfm_get_opup(ofport->cfm);
2978 cfm_run(ofport->cfm);
2979 enable = enable && !cfm_get_fault(ofport->cfm);
2981 if (cfm_opup >= 0) {
2982 enable = enable && cfm_opup;
2986 if (ofport->bundle) {
2987 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
2988 if (carrier_changed) {
2989 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
2993 if (ofport->may_enable != enable) {
2994 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2996 if (ofproto->has_bundle_action) {
2997 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3001 ofport->may_enable = enable;
3005 port_wait(struct ofport_dpif *ofport)
3008 cfm_wait(ofport->cfm);
3013 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3014 struct ofproto_port *ofproto_port)
3016 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3017 struct dpif_port dpif_port;
3020 if (sset_contains(&ofproto->ghost_ports, devname)) {
3021 const char *type = netdev_get_type_from_name(devname);
3023 /* We may be called before ofproto->up.port_by_name is populated with
3024 * the appropriate ofport. For this reason, we must get the name and
3025 * type from the netdev layer directly. */
3027 const struct ofport *ofport;
3029 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3030 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3031 ofproto_port->name = xstrdup(devname);
3032 ofproto_port->type = xstrdup(type);
3038 if (!sset_contains(&ofproto->ports, devname)) {
3041 error = dpif_port_query_by_name(ofproto->backer->dpif,
3042 devname, &dpif_port);
3044 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3050 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3052 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3053 const char *dp_port_name = netdev_vport_get_dpif_port(netdev);
3054 const char *devname = netdev_get_name(netdev);
3056 if (netdev_vport_is_patch(netdev)) {
3057 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3061 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3062 uint32_t port_no = UINT32_MAX;
3065 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3069 if (netdev_get_tunnel_config(netdev)) {
3070 simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no);
3074 if (netdev_get_tunnel_config(netdev)) {
3075 sset_add(&ofproto->ghost_ports, devname);
3077 sset_add(&ofproto->ports, devname);
3083 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
3085 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3086 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3093 sset_find_and_delete(&ofproto->ghost_ports,
3094 netdev_get_name(ofport->up.netdev));
3095 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3096 if (!ofport->tnl_port) {
3097 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3099 /* The caller is going to close ofport->up.netdev. If this is a
3100 * bonded port, then the bond is using that netdev, so remove it
3101 * from the bond. The client will need to reconfigure everything
3102 * after deleting ports, so then the slave will get re-added. */
3103 bundle_remove(&ofport->up);
3110 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3112 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3115 error = netdev_get_stats(ofport->up.netdev, stats);
3117 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3118 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3120 /* ofproto->stats.tx_packets represents packets that we created
3121 * internally and sent to some port (e.g. packets sent with
3122 * send_packet()). Account for them as if they had come from
3123 * OFPP_LOCAL and got forwarded. */
3125 if (stats->rx_packets != UINT64_MAX) {
3126 stats->rx_packets += ofproto->stats.tx_packets;
3129 if (stats->rx_bytes != UINT64_MAX) {
3130 stats->rx_bytes += ofproto->stats.tx_bytes;
3133 /* ofproto->stats.rx_packets represents packets that were received on
3134 * some port and we processed internally and dropped (e.g. STP).
3135 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3137 if (stats->tx_packets != UINT64_MAX) {
3138 stats->tx_packets += ofproto->stats.rx_packets;
3141 if (stats->tx_bytes != UINT64_MAX) {
3142 stats->tx_bytes += ofproto->stats.rx_bytes;
3149 /* Account packets for LOCAL port. */
3151 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
3152 size_t tx_size, size_t rx_size)
3154 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3157 ofproto->stats.rx_packets++;
3158 ofproto->stats.rx_bytes += rx_size;
3161 ofproto->stats.tx_packets++;
3162 ofproto->stats.tx_bytes += tx_size;
3166 struct port_dump_state {
3171 struct ofproto_port port;
3176 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3178 *statep = xzalloc(sizeof(struct port_dump_state));
3183 port_dump_next(const struct ofproto *ofproto_, void *state_,
3184 struct ofproto_port *port)
3186 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3187 struct port_dump_state *state = state_;
3188 const struct sset *sset;
3189 struct sset_node *node;
3191 if (state->has_port) {
3192 ofproto_port_destroy(&state->port);
3193 state->has_port = false;
3195 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3196 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3199 error = port_query_by_name(ofproto_, node->name, &state->port);
3201 *port = state->port;
3202 state->has_port = true;
3204 } else if (error != ENODEV) {
3209 if (!state->ghost) {
3210 state->ghost = true;
3213 return port_dump_next(ofproto_, state_, port);
3220 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3222 struct port_dump_state *state = state_;
3224 if (state->has_port) {
3225 ofproto_port_destroy(&state->port);
3232 port_poll(const struct ofproto *ofproto_, char **devnamep)
3234 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3236 if (ofproto->port_poll_errno) {
3237 int error = ofproto->port_poll_errno;
3238 ofproto->port_poll_errno = 0;
3242 if (sset_is_empty(&ofproto->port_poll_set)) {
3246 *devnamep = sset_pop(&ofproto->port_poll_set);
3251 port_poll_wait(const struct ofproto *ofproto_)
3253 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3254 dpif_port_poll_wait(ofproto->backer->dpif);
3258 port_is_lacp_current(const struct ofport *ofport_)
3260 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3261 return (ofport->bundle && ofport->bundle->lacp
3262 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3266 /* Upcall handling. */
3268 /* Flow miss batching.
3270 * Some dpifs implement operations faster when you hand them off in a batch.
3271 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3272 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3273 * more packets, plus possibly installing the flow in the dpif.
3275 * So far we only batch the operations that affect flow setup time the most.
3276 * It's possible to batch more than that, but the benefit might be minimal. */
3278 struct hmap_node hmap_node;
3279 struct ofproto_dpif *ofproto;
3281 enum odp_key_fitness key_fitness;
3282 const struct nlattr *key;
3284 struct initial_vals initial_vals;
3285 struct list packets;
3286 enum dpif_upcall_type upcall_type;
3287 uint32_t odp_in_port;
3290 struct flow_miss_op {
3291 struct dpif_op dpif_op;
3292 void *garbage; /* Pointer to pass to free(), NULL if none. */
3293 uint64_t stub[1024 / 8]; /* Temporary buffer. */
3296 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3297 * OpenFlow controller as necessary according to their individual
3298 * configurations. */
3300 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3301 const struct flow *flow)
3303 struct ofputil_packet_in pin;
3305 pin.packet = packet->data;
3306 pin.packet_len = packet->size;
3307 pin.reason = OFPR_NO_MATCH;
3308 pin.controller_id = 0;
3313 pin.send_len = 0; /* not used for flow table misses */
3315 flow_get_metadata(flow, &pin.fmd);
3317 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3320 static enum slow_path_reason
3321 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3322 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
3326 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3328 cfm_process_heartbeat(ofport->cfm, packet);
3331 } else if (ofport->bundle && ofport->bundle->lacp
3332 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3334 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3337 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3339 stp_process_packet(ofport, packet);
3347 static struct flow_miss *
3348 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3349 const struct flow *flow, uint32_t hash)
3351 struct flow_miss *miss;
3353 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3354 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3362 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3363 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3364 * 'miss' is associated with a subfacet the caller must also initialize the
3365 * returned op->subfacet, and if anything needs to be freed after processing
3366 * the op, the caller must initialize op->garbage also. */
3368 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3369 struct flow_miss_op *op)
3371 if (miss->flow.vlan_tci != miss->initial_vals.vlan_tci) {
3372 /* This packet was received on a VLAN splinter port. We
3373 * added a VLAN to the packet to make the packet resemble
3374 * the flow, but the actions were composed assuming that
3375 * the packet contained no VLAN. So, we must remove the
3376 * VLAN header from the packet before trying to execute the
3378 eth_pop_vlan(packet);
3382 op->dpif_op.type = DPIF_OP_EXECUTE;
3383 op->dpif_op.u.execute.key = miss->key;
3384 op->dpif_op.u.execute.key_len = miss->key_len;
3385 op->dpif_op.u.execute.packet = packet;
3388 /* Helper for handle_flow_miss_without_facet() and
3389 * handle_flow_miss_with_facet(). */
3391 handle_flow_miss_common(struct rule_dpif *rule,
3392 struct ofpbuf *packet, const struct flow *flow)
3394 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3396 ofproto->n_matches++;
3398 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3400 * Extra-special case for fail-open mode.
3402 * We are in fail-open mode and the packet matched the fail-open
3403 * rule, but we are connected to a controller too. We should send
3404 * the packet up to the controller in the hope that it will try to
3405 * set up a flow and thereby allow us to exit fail-open.
3407 * See the top-level comment in fail-open.c for more information.
3409 send_packet_in_miss(ofproto, packet, flow);
3413 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3414 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3415 * installing a datapath flow. The answer is usually "yes" (a return value of
3416 * true). However, for short flows the cost of bookkeeping is much higher than
3417 * the benefits, so when the datapath holds a large number of flows we impose
3418 * some heuristics to decide which flows are likely to be worth tracking. */
3420 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3421 struct flow_miss *miss, uint32_t hash)
3423 if (!ofproto->governor) {
3426 n_subfacets = hmap_count(&ofproto->subfacets);
3427 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3431 ofproto->governor = governor_create(ofproto->up.name);
3434 return governor_should_install_flow(ofproto->governor, hash,
3435 list_size(&miss->packets));
3438 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3439 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3440 * increment '*n_ops'. */
3442 handle_flow_miss_without_facet(struct flow_miss *miss,
3443 struct rule_dpif *rule,
3444 struct flow_miss_op *ops, size_t *n_ops)
3446 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3447 long long int now = time_msec();
3448 struct action_xlate_ctx ctx;
3449 struct ofpbuf *packet;
3451 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3452 struct flow_miss_op *op = &ops[*n_ops];
3453 struct dpif_flow_stats stats;
3454 struct ofpbuf odp_actions;
3456 COVERAGE_INC(facet_suppress);
3458 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3460 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3461 rule_credit_stats(rule, &stats);
3463 action_xlate_ctx_init(&ctx, ofproto, &miss->flow,
3464 &miss->initial_vals, rule, 0, packet);
3465 ctx.resubmit_stats = &stats;
3466 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
3469 if (odp_actions.size) {
3470 struct dpif_execute *execute = &op->dpif_op.u.execute;
3472 init_flow_miss_execute_op(miss, packet, op);
3473 execute->actions = odp_actions.data;
3474 execute->actions_len = odp_actions.size;
3475 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3479 ofpbuf_uninit(&odp_actions);
3484 /* Handles 'miss', which matches 'facet'. May add any required datapath
3485 * operations to 'ops', incrementing '*n_ops' for each new op.
3487 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3488 * This is really important only for new facets: if we just called time_msec()
3489 * here, then the new subfacet or its packets could look (occasionally) as
3490 * though it was used some time after the facet was used. That can make a
3491 * one-packet flow look like it has a nonzero duration, which looks odd in
3492 * e.g. NetFlow statistics. */
3494 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3496 struct flow_miss_op *ops, size_t *n_ops)
3498 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3499 enum subfacet_path want_path;
3500 struct subfacet *subfacet;
3501 struct ofpbuf *packet;
3503 subfacet = subfacet_create(facet, miss, now);
3505 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3506 struct flow_miss_op *op = &ops[*n_ops];
3507 struct dpif_flow_stats stats;
3508 struct ofpbuf odp_actions;
3510 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3512 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3513 if (!subfacet->actions || subfacet->slow) {
3514 subfacet_make_actions(subfacet, packet, &odp_actions);
3517 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3518 subfacet_update_stats(subfacet, &stats);
3520 if (subfacet->actions_len) {
3521 struct dpif_execute *execute = &op->dpif_op.u.execute;
3523 init_flow_miss_execute_op(miss, packet, op);
3524 if (!subfacet->slow) {
3525 execute->actions = subfacet->actions;
3526 execute->actions_len = subfacet->actions_len;
3527 ofpbuf_uninit(&odp_actions);
3529 execute->actions = odp_actions.data;
3530 execute->actions_len = odp_actions.size;
3531 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3536 ofpbuf_uninit(&odp_actions);
3540 want_path = subfacet_want_path(subfacet->slow);
3541 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3542 struct flow_miss_op *op = &ops[(*n_ops)++];
3543 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3545 subfacet->path = want_path;
3548 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3549 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3550 put->key = miss->key;
3551 put->key_len = miss->key_len;
3552 if (want_path == SF_FAST_PATH) {
3553 put->actions = subfacet->actions;
3554 put->actions_len = subfacet->actions_len;
3556 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
3557 op->stub, sizeof op->stub,
3558 &put->actions, &put->actions_len);
3564 /* Handles flow miss 'miss'. May add any required datapath operations
3565 * to 'ops', incrementing '*n_ops' for each new op. */
3567 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3570 struct ofproto_dpif *ofproto = miss->ofproto;
3571 struct facet *facet;
3575 /* The caller must ensure that miss->hmap_node.hash contains
3576 * flow_hash(miss->flow, 0). */
3577 hash = miss->hmap_node.hash;
3579 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3581 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
3583 if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
3584 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
3588 facet = facet_create(rule, &miss->flow, hash);
3593 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3596 static struct drop_key *
3597 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3600 struct drop_key *drop_key;
3602 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3603 &backer->drop_keys) {
3604 if (drop_key->key_len == key_len
3605 && !memcmp(drop_key->key, key, key_len)) {
3613 drop_key_clear(struct dpif_backer *backer)
3615 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3616 struct drop_key *drop_key, *next;
3618 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3621 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3623 if (error && !VLOG_DROP_WARN(&rl)) {
3624 struct ds ds = DS_EMPTY_INITIALIZER;
3625 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3626 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3631 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3632 free(drop_key->key);
3637 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3638 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3639 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3640 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3641 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3642 * 'packet' ingressed.
3644 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3645 * 'flow''s in_port to OFPP_NONE.
3647 * This function does post-processing on data returned from
3648 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3649 * of the upcall processing logic. In particular, if the extracted in_port is
3650 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3651 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3652 * a VLAN header onto 'packet' (if it is nonnull).
3654 * Optionally, if 'initial_vals' is nonnull, sets 'initial_vals->vlan_tci'
3655 * to the VLAN TCI with which the packet was really received, that is, the
3656 * actual VLAN TCI extracted by odp_flow_key_to_flow(). (This differs from
3657 * the value returned in flow->vlan_tci only for packets received on
3658 * VLAN splinters.) Also, if received on an IP tunnel, sets
3659 * 'initial_vals->tunnel_ip_tos' to the tunnel's IP TOS.
3661 * Similarly, this function also includes some logic to help with tunnels. It
3662 * may modify 'flow' as necessary to make the tunneling implementation
3663 * transparent to the upcall processing logic.
3665 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3666 * or some other positive errno if there are other problems. */
3668 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3669 const struct nlattr *key, size_t key_len,
3670 struct flow *flow, enum odp_key_fitness *fitnessp,
3671 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3672 struct initial_vals *initial_vals)
3674 const struct ofport_dpif *port;
3675 enum odp_key_fitness fitness;
3678 fitness = odp_flow_key_to_flow(key, key_len, flow);
3679 if (fitness == ODP_FIT_ERROR) {
3685 initial_vals->vlan_tci = flow->vlan_tci;
3686 initial_vals->tunnel_ip_tos = flow->tunnel.ip_tos;
3690 *odp_in_port = flow->in_port;
3693 if (tnl_port_should_receive(flow)) {
3694 const struct ofport *ofport = tnl_port_receive(flow);
3696 flow->in_port = OFPP_NONE;
3699 port = ofport_dpif_cast(ofport);
3701 /* We can't reproduce 'key' from 'flow'. */
3702 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3704 /* XXX: Since the tunnel module is not scoped per backer, it's
3705 * theoretically possible that we'll receive an ofport belonging to an
3706 * entirely different datapath. In practice, this can't happen because
3707 * no platforms has two separate datapaths which each support
3709 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3711 port = odp_port_to_ofport(backer, flow->in_port);
3713 flow->in_port = OFPP_NONE;
3717 flow->in_port = port->up.ofp_port;
3718 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3720 /* Make the packet resemble the flow, so that it gets sent to
3721 * an OpenFlow controller properly, so that it looks correct
3722 * for sFlow, and so that flow_extract() will get the correct
3723 * vlan_tci if it is called on 'packet'.
3725 * The allocated space inside 'packet' probably also contains
3726 * 'key', that is, both 'packet' and 'key' are probably part of
3727 * a struct dpif_upcall (see the large comment on that
3728 * structure definition), so pushing data on 'packet' is in
3729 * general not a good idea since it could overwrite 'key' or
3730 * free it as a side effect. However, it's OK in this special
3731 * case because we know that 'packet' is inside a Netlink
3732 * attribute: pushing 4 bytes will just overwrite the 4-byte
3733 * "struct nlattr", which is fine since we don't need that
3734 * header anymore. */
3735 eth_push_vlan(packet, flow->vlan_tci);
3737 /* We can't reproduce 'key' from 'flow'. */
3738 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3744 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3749 *fitnessp = fitness;
3755 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3758 struct dpif_upcall *upcall;
3759 struct flow_miss *miss;
3760 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3761 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3762 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3772 /* Construct the to-do list.
3774 * This just amounts to extracting the flow from each packet and sticking
3775 * the packets that have the same flow in the same "flow_miss" structure so
3776 * that we can process them together. */
3779 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3780 struct flow_miss *miss = &misses[n_misses];
3781 struct flow_miss *existing_miss;
3782 struct ofproto_dpif *ofproto;
3783 uint32_t odp_in_port;
3788 error = ofproto_receive(backer, upcall->packet, upcall->key,
3789 upcall->key_len, &flow, &miss->key_fitness,
3790 &ofproto, &odp_in_port, &miss->initial_vals);
3791 if (error == ENODEV) {
3792 struct drop_key *drop_key;
3794 /* Received packet on port for which we couldn't associate
3795 * an ofproto. This can happen if a port is removed while
3796 * traffic is being received. Print a rate-limited message
3797 * in case it happens frequently. Install a drop flow so
3798 * that future packets of the flow are inexpensively dropped
3800 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
3803 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
3805 drop_key = xmalloc(sizeof *drop_key);
3806 drop_key->key = xmemdup(upcall->key, upcall->key_len);
3807 drop_key->key_len = upcall->key_len;
3809 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3810 hash_bytes(drop_key->key, drop_key->key_len, 0));
3811 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3812 drop_key->key, drop_key->key_len, NULL, 0, NULL);
3819 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3820 &flow.tunnel, flow.in_port, &miss->flow);
3822 /* Add other packets to a to-do list. */
3823 hash = flow_hash(&miss->flow, 0);
3824 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
3825 if (!existing_miss) {
3826 hmap_insert(&todo, &miss->hmap_node, hash);
3827 miss->ofproto = ofproto;
3828 miss->key = upcall->key;
3829 miss->key_len = upcall->key_len;
3830 miss->upcall_type = upcall->type;
3831 miss->odp_in_port = odp_in_port;
3832 list_init(&miss->packets);
3836 miss = existing_miss;
3838 list_push_back(&miss->packets, &upcall->packet->list_node);
3841 /* Process each element in the to-do list, constructing the set of
3842 * operations to batch. */
3844 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3845 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3847 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3849 /* Execute batch. */
3850 for (i = 0; i < n_ops; i++) {
3851 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3853 dpif_operate(backer->dpif, dpif_ops, n_ops);
3856 for (i = 0; i < n_ops; i++) {
3857 free(flow_miss_ops[i].garbage);
3859 hmap_destroy(&todo);
3862 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL }
3863 classify_upcall(const struct dpif_upcall *upcall)
3865 union user_action_cookie cookie;
3867 /* First look at the upcall type. */
3868 switch (upcall->type) {
3869 case DPIF_UC_ACTION:
3875 case DPIF_N_UC_TYPES:
3877 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3881 /* "action" upcalls need a closer look. */
3882 if (!upcall->userdata) {
3883 VLOG_WARN_RL(&rl, "action upcall missing cookie");
3886 if (nl_attr_get_size(upcall->userdata) != sizeof(cookie)) {
3887 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
3888 nl_attr_get_size(upcall->userdata));
3891 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof(cookie));
3892 switch (cookie.type) {
3893 case USER_ACTION_COOKIE_SFLOW:
3894 return SFLOW_UPCALL;
3896 case USER_ACTION_COOKIE_SLOW_PATH:
3899 case USER_ACTION_COOKIE_UNSPEC:
3901 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64,
3902 nl_attr_get_u64(upcall->userdata));
3908 handle_sflow_upcall(struct dpif_backer *backer,
3909 const struct dpif_upcall *upcall)
3911 struct ofproto_dpif *ofproto;
3912 union user_action_cookie cookie;
3914 uint32_t odp_in_port;
3916 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3917 &flow, NULL, &ofproto, &odp_in_port, NULL)
3918 || !ofproto->sflow) {
3922 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof(cookie));
3923 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3924 odp_in_port, &cookie);
3928 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
3930 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3931 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3932 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3937 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
3940 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3941 struct dpif_upcall *upcall = &misses[n_misses];
3942 struct ofpbuf *buf = &miss_bufs[n_misses];
3945 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3946 sizeof miss_buf_stubs[n_misses]);
3947 error = dpif_recv(backer->dpif, upcall, buf);
3953 switch (classify_upcall(upcall)) {
3955 /* Handle it later. */
3960 handle_sflow_upcall(backer, upcall);
3970 /* Handle deferred MISS_UPCALL processing. */
3971 handle_miss_upcalls(backer, misses, n_misses);
3972 for (i = 0; i < n_misses; i++) {
3973 ofpbuf_uninit(&miss_bufs[i]);
3979 /* Flow expiration. */
3981 static int subfacet_max_idle(const struct ofproto_dpif *);
3982 static void update_stats(struct dpif_backer *);
3983 static void rule_expire(struct rule_dpif *);
3984 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
3986 /* This function is called periodically by run(). Its job is to collect
3987 * updates for the flows that have been installed into the datapath, most
3988 * importantly when they last were used, and then use that information to
3989 * expire flows that have not been used recently.
3991 * Returns the number of milliseconds after which it should be called again. */
3993 expire(struct dpif_backer *backer)
3995 struct ofproto_dpif *ofproto;
3996 int max_idle = INT32_MAX;
3998 /* Periodically clear out the drop keys in an effort to keep them
3999 * relatively few. */
4000 drop_key_clear(backer);
4002 /* Update stats for each flow in the backer. */
4003 update_stats(backer);
4005 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4006 struct rule *rule, *next_rule;
4009 if (ofproto->backer != backer) {
4013 /* Expire subfacets that have been idle too long. */
4014 dp_max_idle = subfacet_max_idle(ofproto);
4015 expire_subfacets(ofproto, dp_max_idle);
4017 max_idle = MIN(max_idle, dp_max_idle);
4019 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4021 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4022 &ofproto->up.expirable) {
4023 rule_expire(rule_dpif_cast(rule));
4026 /* All outstanding data in existing flows has been accounted, so it's a
4027 * good time to do bond rebalancing. */
4028 if (ofproto->has_bonded_bundles) {
4029 struct ofbundle *bundle;
4031 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4033 bond_rebalance(bundle->bond, &backer->revalidate_set);
4039 return MIN(max_idle, 1000);
4042 /* Updates flow table statistics given that the datapath just reported 'stats'
4043 * as 'subfacet''s statistics. */
4045 update_subfacet_stats(struct subfacet *subfacet,
4046 const struct dpif_flow_stats *stats)
4048 struct facet *facet = subfacet->facet;
4050 if (stats->n_packets >= subfacet->dp_packet_count) {
4051 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
4052 facet->packet_count += extra;
4054 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4057 if (stats->n_bytes >= subfacet->dp_byte_count) {
4058 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
4060 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4063 subfacet->dp_packet_count = stats->n_packets;
4064 subfacet->dp_byte_count = stats->n_bytes;
4066 facet->tcp_flags |= stats->tcp_flags;
4068 subfacet_update_time(subfacet, stats->used);
4069 if (facet->accounted_bytes < facet->byte_count) {
4071 facet_account(facet);
4072 facet->accounted_bytes = facet->byte_count;
4074 facet_push_stats(facet);
4077 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4078 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4080 delete_unexpected_flow(struct ofproto_dpif *ofproto,
4081 const struct nlattr *key, size_t key_len)
4083 if (!VLOG_DROP_WARN(&rl)) {
4087 odp_flow_key_format(key, key_len, &s);
4088 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
4092 COVERAGE_INC(facet_unexpected);
4093 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
4096 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4098 * This function also pushes statistics updates to rules which each facet
4099 * resubmits into. Generally these statistics will be accurate. However, if a
4100 * facet changes the rule it resubmits into at some time in between
4101 * update_stats() runs, it is possible that statistics accrued to the
4102 * old rule will be incorrectly attributed to the new rule. This could be
4103 * avoided by calling update_stats() whenever rules are created or
4104 * deleted. However, the performance impact of making so many calls to the
4105 * datapath do not justify the benefit of having perfectly accurate statistics.
4108 update_stats(struct dpif_backer *backer)
4110 const struct dpif_flow_stats *stats;
4111 struct dpif_flow_dump dump;
4112 const struct nlattr *key;
4115 dpif_flow_dump_start(&dump, backer->dpif);
4116 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4118 struct subfacet *subfacet;
4119 struct ofproto_dpif *ofproto;
4120 struct ofport_dpif *ofport;
4123 if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto,
4128 ofport = get_ofp_port(ofproto, flow.in_port);
4129 if (ofport && ofport->tnl_port) {
4130 netdev_vport_inc_rx(ofport->up.netdev, stats);
4133 key_hash = odp_flow_key_hash(key, key_len);
4134 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
4135 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4137 update_subfacet_stats(subfacet, stats);
4141 /* Stats are updated per-packet. */
4144 case SF_NOT_INSTALLED:
4146 delete_unexpected_flow(ofproto, key, key_len);
4150 dpif_flow_dump_done(&dump);
4153 /* Calculates and returns the number of milliseconds of idle time after which
4154 * subfacets should expire from the datapath. When a subfacet expires, we fold
4155 * its statistics into its facet, and when a facet's last subfacet expires, we
4156 * fold its statistic into its rule. */
4158 subfacet_max_idle(const struct ofproto_dpif *ofproto)
4161 * Idle time histogram.
4163 * Most of the time a switch has a relatively small number of subfacets.
4164 * When this is the case we might as well keep statistics for all of them
4165 * in userspace and to cache them in the kernel datapath for performance as
4168 * As the number of subfacets increases, the memory required to maintain
4169 * statistics about them in userspace and in the kernel becomes
4170 * significant. However, with a large number of subfacets it is likely
4171 * that only a few of them are "heavy hitters" that consume a large amount
4172 * of bandwidth. At this point, only heavy hitters are worth caching in
4173 * the kernel and maintaining in userspaces; other subfacets we can
4176 * The technique used to compute the idle time is to build a histogram with
4177 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4178 * that is installed in the kernel gets dropped in the appropriate bucket.
4179 * After the histogram has been built, we compute the cutoff so that only
4180 * the most-recently-used 1% of subfacets (but at least
4181 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
4182 * the most-recently-used bucket of subfacets is kept, so actually an
4183 * arbitrary number of subfacets can be kept in any given expiration run
4184 * (though the next run will delete most of those unless they receive
4187 * This requires a second pass through the subfacets, in addition to the
4188 * pass made by update_stats(), because the former function never looks at
4189 * uninstallable subfacets.
4191 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4192 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4193 int buckets[N_BUCKETS] = { 0 };
4194 int total, subtotal, bucket;
4195 struct subfacet *subfacet;
4199 total = hmap_count(&ofproto->subfacets);
4200 if (total <= ofproto->up.flow_eviction_threshold) {
4201 return N_BUCKETS * BUCKET_WIDTH;
4204 /* Build histogram. */
4206 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
4207 long long int idle = now - subfacet->used;
4208 int bucket = (idle <= 0 ? 0
4209 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4210 : (unsigned int) idle / BUCKET_WIDTH);
4214 /* Find the first bucket whose flows should be expired. */
4215 subtotal = bucket = 0;
4217 subtotal += buckets[bucket++];
4218 } while (bucket < N_BUCKETS &&
4219 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
4221 if (VLOG_IS_DBG_ENABLED()) {
4225 ds_put_cstr(&s, "keep");
4226 for (i = 0; i < N_BUCKETS; i++) {
4228 ds_put_cstr(&s, ", drop");
4231 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4234 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
4238 return bucket * BUCKET_WIDTH;
4242 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
4244 /* Cutoff time for most flows. */
4245 long long int normal_cutoff = time_msec() - dp_max_idle;
4247 /* We really want to keep flows for special protocols around, so use a more
4248 * conservative cutoff. */
4249 long long int special_cutoff = time_msec() - 10000;
4251 struct subfacet *subfacet, *next_subfacet;
4252 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4256 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4257 &ofproto->subfacets) {
4258 long long int cutoff;
4260 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)
4263 if (subfacet->used < cutoff) {
4264 if (subfacet->path != SF_NOT_INSTALLED) {
4265 batch[n_batch++] = subfacet;
4266 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4267 subfacet_destroy_batch(ofproto, batch, n_batch);
4271 subfacet_destroy(subfacet);
4277 subfacet_destroy_batch(ofproto, batch, n_batch);
4281 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4282 * then delete it entirely. */
4284 rule_expire(struct rule_dpif *rule)
4286 struct facet *facet, *next_facet;
4290 if (rule->up.pending) {
4291 /* We'll have to expire it later. */
4295 /* Has 'rule' expired? */
4297 if (rule->up.hard_timeout
4298 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4299 reason = OFPRR_HARD_TIMEOUT;
4300 } else if (rule->up.idle_timeout
4301 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4302 reason = OFPRR_IDLE_TIMEOUT;
4307 COVERAGE_INC(ofproto_dpif_expired);
4309 /* Update stats. (This is a no-op if the rule expired due to an idle
4310 * timeout, because that only happens when the rule has no facets left.) */
4311 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4312 facet_remove(facet);
4315 /* Get rid of the rule. */
4316 ofproto_rule_expire(&rule->up, reason);
4321 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
4323 * The caller must already have determined that no facet with an identical
4324 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
4325 * the ofproto's classifier table.
4327 * 'hash' must be the return value of flow_hash(flow, 0).
4329 * The facet will initially have no subfacets. The caller should create (at
4330 * least) one subfacet with subfacet_create(). */
4331 static struct facet *
4332 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
4334 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4335 struct facet *facet;
4337 facet = xzalloc(sizeof *facet);
4338 facet->used = time_msec();
4339 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4340 list_push_back(&rule->facets, &facet->list_node);
4342 facet->flow = *flow;
4343 list_init(&facet->subfacets);
4344 netflow_flow_init(&facet->nf_flow);
4345 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4351 facet_free(struct facet *facet)
4356 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4357 * 'packet', which arrived on 'in_port'. */
4359 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4360 const struct nlattr *odp_actions, size_t actions_len,
4361 struct ofpbuf *packet)
4363 struct odputil_keybuf keybuf;
4367 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4368 odp_flow_key_from_flow(&key, flow,
4369 ofp_port_to_odp_port(ofproto, flow->in_port));
4371 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4372 odp_actions, actions_len, packet);
4376 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4378 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4379 * rule's statistics, via subfacet_uninstall().
4381 * - Removes 'facet' from its rule and from ofproto->facets.
4384 facet_remove(struct facet *facet)
4386 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4387 struct subfacet *subfacet, *next_subfacet;
4389 ovs_assert(!list_is_empty(&facet->subfacets));
4391 /* First uninstall all of the subfacets to get final statistics. */
4392 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4393 subfacet_uninstall(subfacet);
4396 /* Flush the final stats to the rule.
4398 * This might require us to have at least one subfacet around so that we
4399 * can use its actions for accounting in facet_account(), which is why we
4400 * have uninstalled but not yet destroyed the subfacets. */
4401 facet_flush_stats(facet);
4403 /* Now we're really all done so destroy everything. */
4404 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4405 &facet->subfacets) {
4406 subfacet_destroy__(subfacet);
4408 hmap_remove(&ofproto->facets, &facet->hmap_node);
4409 list_remove(&facet->list_node);
4413 /* Feed information from 'facet' back into the learning table to keep it in
4414 * sync with what is actually flowing through the datapath. */
4416 facet_learn(struct facet *facet)
4418 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4419 struct subfacet *subfacet= CONTAINER_OF(list_front(&facet->subfacets),
4420 struct subfacet, list_node);
4421 struct action_xlate_ctx ctx;
4423 if (!facet->has_learn
4424 && !facet->has_normal
4425 && (!facet->has_fin_timeout
4426 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4430 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4431 &subfacet->initial_vals,
4432 facet->rule, facet->tcp_flags, NULL);
4433 ctx.may_learn = true;
4434 xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
4435 facet->rule->up.ofpacts_len);
4439 facet_account(struct facet *facet)
4441 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4442 struct subfacet *subfacet;
4443 const struct nlattr *a;
4448 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
4451 n_bytes = facet->byte_count - facet->accounted_bytes;
4453 /* This loop feeds byte counters to bond_account() for rebalancing to use
4454 * as a basis. We also need to track the actual VLAN on which the packet
4455 * is going to be sent to ensure that it matches the one passed to
4456 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4459 * We use the actions from an arbitrary subfacet because they should all
4460 * be equally valid for our purpose. */
4461 subfacet = CONTAINER_OF(list_front(&facet->subfacets),
4462 struct subfacet, list_node);
4463 vlan_tci = facet->flow.vlan_tci;
4464 NL_ATTR_FOR_EACH_UNSAFE (a, left,
4465 subfacet->actions, subfacet->actions_len) {
4466 const struct ovs_action_push_vlan *vlan;
4467 struct ofport_dpif *port;
4469 switch (nl_attr_type(a)) {
4470 case OVS_ACTION_ATTR_OUTPUT:
4471 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4472 if (port && port->bundle && port->bundle->bond) {
4473 bond_account(port->bundle->bond, &facet->flow,
4474 vlan_tci_to_vid(vlan_tci), n_bytes);
4478 case OVS_ACTION_ATTR_POP_VLAN:
4479 vlan_tci = htons(0);
4482 case OVS_ACTION_ATTR_PUSH_VLAN:
4483 vlan = nl_attr_get(a);
4484 vlan_tci = vlan->vlan_tci;
4490 /* Returns true if the only action for 'facet' is to send to the controller.
4491 * (We don't report NetFlow expiration messages for such facets because they
4492 * are just part of the control logic for the network, not real traffic). */
4494 facet_is_controller_flow(struct facet *facet)
4497 const struct rule *rule = &facet->rule->up;
4498 const struct ofpact *ofpacts = rule->ofpacts;
4499 size_t ofpacts_len = rule->ofpacts_len;
4501 if (ofpacts_len > 0 &&
4502 ofpacts->type == OFPACT_CONTROLLER &&
4503 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4510 /* Folds all of 'facet''s statistics into its rule. Also updates the
4511 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4512 * 'facet''s statistics in the datapath should have been zeroed and folded into
4513 * its packet and byte counts before this function is called. */
4515 facet_flush_stats(struct facet *facet)
4517 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4518 struct subfacet *subfacet;
4520 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4521 ovs_assert(!subfacet->dp_byte_count);
4522 ovs_assert(!subfacet->dp_packet_count);
4525 facet_push_stats(facet);
4526 if (facet->accounted_bytes < facet->byte_count) {
4527 facet_account(facet);
4528 facet->accounted_bytes = facet->byte_count;
4531 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4532 struct ofexpired expired;
4533 expired.flow = facet->flow;
4534 expired.packet_count = facet->packet_count;
4535 expired.byte_count = facet->byte_count;
4536 expired.used = facet->used;
4537 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4540 facet->rule->packet_count += facet->packet_count;
4541 facet->rule->byte_count += facet->byte_count;
4543 /* Reset counters to prevent double counting if 'facet' ever gets
4545 facet_reset_counters(facet);
4547 netflow_flow_clear(&facet->nf_flow);
4548 facet->tcp_flags = 0;
4551 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4552 * Returns it if found, otherwise a null pointer.
4554 * 'hash' must be the return value of flow_hash(flow, 0).
4556 * The returned facet might need revalidation; use facet_lookup_valid()
4557 * instead if that is important. */
4558 static struct facet *
4559 facet_find(struct ofproto_dpif *ofproto,
4560 const struct flow *flow, uint32_t hash)
4562 struct facet *facet;
4564 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4565 if (flow_equal(flow, &facet->flow)) {
4573 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4574 * Returns it if found, otherwise a null pointer.
4576 * 'hash' must be the return value of flow_hash(flow, 0).
4578 * The returned facet is guaranteed to be valid. */
4579 static struct facet *
4580 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4583 struct facet *facet;
4585 facet = facet_find(ofproto, flow, hash);
4587 && (ofproto->backer->need_revalidate
4588 || tag_set_intersects(&ofproto->backer->revalidate_set,
4590 facet_revalidate(facet);
4592 /* facet_revalidate() may have destroyed 'facet'. */
4593 facet = facet_find(ofproto, flow, hash);
4600 subfacet_path_to_string(enum subfacet_path path)
4603 case SF_NOT_INSTALLED:
4604 return "not installed";
4606 return "in fast path";
4608 return "in slow path";
4614 /* Returns the path in which a subfacet should be installed if its 'slow'
4615 * member has the specified value. */
4616 static enum subfacet_path
4617 subfacet_want_path(enum slow_path_reason slow)
4619 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
4622 /* Returns true if 'subfacet' needs to have its datapath flow updated,
4623 * supposing that its actions have been recalculated as 'want_actions' and that
4624 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
4626 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
4627 const struct ofpbuf *want_actions)
4629 enum subfacet_path want_path = subfacet_want_path(slow);
4630 return (want_path != subfacet->path
4631 || (want_path == SF_FAST_PATH
4632 && (subfacet->actions_len != want_actions->size
4633 || memcmp(subfacet->actions, want_actions->data,
4634 subfacet->actions_len))));
4638 facet_check_consistency(struct facet *facet)
4640 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4642 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4644 uint64_t odp_actions_stub[1024 / 8];
4645 struct ofpbuf odp_actions;
4647 struct rule_dpif *rule;
4648 struct subfacet *subfacet;
4649 bool may_log = false;
4652 /* Check the rule for consistency. */
4653 rule = rule_dpif_lookup(ofproto, &facet->flow);
4654 ok = rule == facet->rule;
4656 may_log = !VLOG_DROP_WARN(&rl);
4661 flow_format(&s, &facet->flow);
4662 ds_put_format(&s, ": facet associated with wrong rule (was "
4663 "table=%"PRIu8",", facet->rule->up.table_id);
4664 cls_rule_format(&facet->rule->up.cr, &s);
4665 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4667 cls_rule_format(&rule->up.cr, &s);
4668 ds_put_char(&s, ')');
4670 VLOG_WARN("%s", ds_cstr(&s));
4675 /* Check the datapath actions for consistency. */
4676 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4677 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4678 enum subfacet_path want_path;
4679 struct action_xlate_ctx ctx;
4682 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4683 &subfacet->initial_vals, rule, 0, NULL);
4684 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
4687 if (subfacet->path == SF_NOT_INSTALLED) {
4688 /* This only happens if the datapath reported an error when we
4689 * tried to install the flow. Don't flag another error here. */
4693 want_path = subfacet_want_path(subfacet->slow);
4694 if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) {
4695 /* The actions for slow-path flows may legitimately vary from one
4696 * packet to the next. We're done. */
4700 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
4704 /* Inconsistency! */
4706 may_log = !VLOG_DROP_WARN(&rl);
4710 /* Rate-limited, skip reporting. */
4715 odp_flow_key_format(subfacet->key, subfacet->key_len, &s);
4717 ds_put_cstr(&s, ": inconsistency in subfacet");
4718 if (want_path != subfacet->path) {
4719 enum odp_key_fitness fitness = subfacet->key_fitness;
4721 ds_put_format(&s, " (%s, fitness=%s)",
4722 subfacet_path_to_string(subfacet->path),
4723 odp_key_fitness_to_string(fitness));
4724 ds_put_format(&s, " (should have been %s)",
4725 subfacet_path_to_string(want_path));
4726 } else if (want_path == SF_FAST_PATH) {
4727 ds_put_cstr(&s, " (actions were: ");
4728 format_odp_actions(&s, subfacet->actions,
4729 subfacet->actions_len);
4730 ds_put_cstr(&s, ") (correct actions: ");
4731 format_odp_actions(&s, odp_actions.data, odp_actions.size);
4732 ds_put_char(&s, ')');
4734 ds_put_cstr(&s, " (actions: ");
4735 format_odp_actions(&s, subfacet->actions,
4736 subfacet->actions_len);
4737 ds_put_char(&s, ')');
4739 VLOG_WARN("%s", ds_cstr(&s));
4742 ofpbuf_uninit(&odp_actions);
4747 /* Re-searches the classifier for 'facet':
4749 * - If the rule found is different from 'facet''s current rule, moves
4750 * 'facet' to the new rule and recompiles its actions.
4752 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4753 * where it is and recompiles its actions anyway.
4755 * - If any of 'facet''s subfacets correspond to a new flow according to
4756 * ofproto_receive(), 'facet' is removed. */
4758 facet_revalidate(struct facet *facet)
4760 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4762 struct nlattr *odp_actions;
4765 struct actions *new_actions;
4767 struct action_xlate_ctx ctx;
4768 uint64_t odp_actions_stub[1024 / 8];
4769 struct ofpbuf odp_actions;
4771 struct rule_dpif *new_rule;
4772 struct subfacet *subfacet;
4775 COVERAGE_INC(facet_revalidate);
4777 /* Check that child subfacets still correspond to this facet. Tunnel
4778 * configuration changes could cause a subfacet's OpenFlow in_port to
4780 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4781 struct ofproto_dpif *recv_ofproto;
4782 struct flow recv_flow;
4785 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
4786 subfacet->key_len, &recv_flow, NULL,
4787 &recv_ofproto, NULL, NULL);
4789 || recv_ofproto != ofproto
4790 || memcmp(&recv_flow, &facet->flow, sizeof recv_flow)) {
4791 facet_remove(facet);
4796 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
4798 /* Calculate new datapath actions.
4800 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4801 * emit a NetFlow expiration and, if so, we need to have the old state
4802 * around to properly compose it. */
4804 /* If the datapath actions changed or the installability changed,
4805 * then we need to talk to the datapath. */
4808 memset(&ctx, 0, sizeof ctx);
4809 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4810 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4811 enum slow_path_reason slow;
4813 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4814 &subfacet->initial_vals, new_rule, 0, NULL);
4815 xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len,
4818 slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4819 if (subfacet_should_install(subfacet, slow, &odp_actions)) {
4820 struct dpif_flow_stats stats;
4822 subfacet_install(subfacet,
4823 odp_actions.data, odp_actions.size, &stats, slow);
4824 subfacet_update_stats(subfacet, &stats);
4827 new_actions = xcalloc(list_size(&facet->subfacets),
4828 sizeof *new_actions);
4830 new_actions[i].odp_actions = xmemdup(odp_actions.data,
4832 new_actions[i].actions_len = odp_actions.size;
4837 ofpbuf_uninit(&odp_actions);
4840 facet_flush_stats(facet);
4843 /* Update 'facet' now that we've taken care of all the old state. */
4844 facet->tags = ctx.tags;
4845 facet->nf_flow.output_iface = ctx.nf_output_iface;
4846 facet->has_learn = ctx.has_learn;
4847 facet->has_normal = ctx.has_normal;
4848 facet->has_fin_timeout = ctx.has_fin_timeout;
4849 facet->mirrors = ctx.mirrors;
4852 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4853 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4855 if (new_actions && new_actions[i].odp_actions) {
4856 free(subfacet->actions);
4857 subfacet->actions = new_actions[i].odp_actions;
4858 subfacet->actions_len = new_actions[i].actions_len;
4864 if (facet->rule != new_rule) {
4865 COVERAGE_INC(facet_changed_rule);
4866 list_remove(&facet->list_node);
4867 list_push_back(&new_rule->facets, &facet->list_node);
4868 facet->rule = new_rule;
4869 facet->used = new_rule->up.created;
4870 facet->prev_used = facet->used;
4874 /* Updates 'facet''s used time. Caller is responsible for calling
4875 * facet_push_stats() to update the flows which 'facet' resubmits into. */
4877 facet_update_time(struct facet *facet, long long int used)
4879 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4880 if (used > facet->used) {
4882 ofproto_rule_update_used(&facet->rule->up, used);
4883 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
4888 facet_reset_counters(struct facet *facet)
4890 facet->packet_count = 0;
4891 facet->byte_count = 0;
4892 facet->prev_packet_count = 0;
4893 facet->prev_byte_count = 0;
4894 facet->accounted_bytes = 0;
4898 facet_push_stats(struct facet *facet)
4900 struct dpif_flow_stats stats;
4902 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4903 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4904 ovs_assert(facet->used >= facet->prev_used);
4906 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4907 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4908 stats.used = facet->used;
4909 stats.tcp_flags = 0;
4911 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
4912 facet->prev_packet_count = facet->packet_count;
4913 facet->prev_byte_count = facet->byte_count;
4914 facet->prev_used = facet->used;
4916 flow_push_stats(facet, &stats);
4918 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
4919 facet->mirrors, stats.n_packets, stats.n_bytes);
4924 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4926 rule->packet_count += stats->n_packets;
4927 rule->byte_count += stats->n_bytes;
4928 ofproto_rule_update_used(&rule->up, stats->used);
4931 /* Pushes flow statistics to the rules which 'facet->flow' resubmits
4932 * into given 'facet->rule''s actions and mirrors. */
4934 flow_push_stats(struct facet *facet, const struct dpif_flow_stats *stats)
4936 struct rule_dpif *rule = facet->rule;
4937 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4938 struct subfacet *subfacet = CONTAINER_OF(list_front(&facet->subfacets),
4939 struct subfacet, list_node);
4940 struct action_xlate_ctx ctx;
4942 ofproto_rule_update_used(&rule->up, stats->used);
4944 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4945 &subfacet->initial_vals, rule, 0, NULL);
4946 ctx.resubmit_stats = stats;
4947 xlate_actions_for_side_effects(&ctx, rule->up.ofpacts,
4948 rule->up.ofpacts_len);
4953 static struct subfacet *
4954 subfacet_find(struct ofproto_dpif *ofproto,
4955 const struct nlattr *key, size_t key_len, uint32_t key_hash)
4957 struct subfacet *subfacet;
4959 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4960 &ofproto->subfacets) {
4961 if (subfacet->key_len == key_len
4962 && !memcmp(key, subfacet->key, key_len)) {
4970 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
4971 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
4972 * existing subfacet if there is one, otherwise creates and returns a
4975 * If the returned subfacet is new, then subfacet->actions will be NULL, in
4976 * which case the caller must populate the actions with
4977 * subfacet_make_actions(). */
4978 static struct subfacet *
4979 subfacet_create(struct facet *facet, struct flow_miss *miss,
4982 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4983 enum odp_key_fitness key_fitness = miss->key_fitness;
4984 const struct nlattr *key = miss->key;
4985 size_t key_len = miss->key_len;
4987 struct subfacet *subfacet;
4989 key_hash = odp_flow_key_hash(key, key_len);
4991 if (list_is_empty(&facet->subfacets)) {
4992 subfacet = &facet->one_subfacet;
4994 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
4996 if (subfacet->facet == facet) {
5000 /* This shouldn't happen. */
5001 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
5002 subfacet_destroy(subfacet);
5005 subfacet = xmalloc(sizeof *subfacet);
5008 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
5009 list_push_back(&facet->subfacets, &subfacet->list_node);
5010 subfacet->facet = facet;
5011 subfacet->key_fitness = key_fitness;
5012 subfacet->key = xmemdup(key, key_len);
5013 subfacet->key_len = key_len;
5014 subfacet->used = now;
5015 subfacet->dp_packet_count = 0;
5016 subfacet->dp_byte_count = 0;
5017 subfacet->actions_len = 0;
5018 subfacet->actions = NULL;
5019 subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE
5022 subfacet->path = SF_NOT_INSTALLED;
5023 subfacet->initial_vals = miss->initial_vals;
5024 subfacet->odp_in_port = miss->odp_in_port;
5029 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
5030 * its facet within 'ofproto', and frees it. */
5032 subfacet_destroy__(struct subfacet *subfacet)
5034 struct facet *facet = subfacet->facet;
5035 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5037 subfacet_uninstall(subfacet);
5038 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
5039 list_remove(&subfacet->list_node);
5040 free(subfacet->key);
5041 free(subfacet->actions);
5042 if (subfacet != &facet->one_subfacet) {
5047 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5048 * last remaining subfacet in its facet destroys the facet too. */
5050 subfacet_destroy(struct subfacet *subfacet)
5052 struct facet *facet = subfacet->facet;
5054 if (list_is_singleton(&facet->subfacets)) {
5055 /* facet_remove() needs at least one subfacet (it will remove it). */
5056 facet_remove(facet);
5058 subfacet_destroy__(subfacet);
5063 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
5064 struct subfacet **subfacets, int n)
5066 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5067 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5068 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5071 for (i = 0; i < n; i++) {
5072 ops[i].type = DPIF_OP_FLOW_DEL;
5073 ops[i].u.flow_del.key = subfacets[i]->key;
5074 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
5075 ops[i].u.flow_del.stats = &stats[i];
5079 dpif_operate(ofproto->backer->dpif, opsp, n);
5080 for (i = 0; i < n; i++) {
5081 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5082 subfacets[i]->path = SF_NOT_INSTALLED;
5083 subfacet_destroy(subfacets[i]);
5087 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
5088 * Translates the actions into 'odp_actions', which the caller must have
5089 * initialized and is responsible for uninitializing. */
5091 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
5092 struct ofpbuf *odp_actions)
5094 struct facet *facet = subfacet->facet;
5095 struct rule_dpif *rule = facet->rule;
5096 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5098 struct action_xlate_ctx ctx;
5100 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
5101 &subfacet->initial_vals, rule, 0, packet);
5102 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions);
5103 facet->tags = ctx.tags;
5104 facet->has_learn = ctx.has_learn;
5105 facet->has_normal = ctx.has_normal;
5106 facet->has_fin_timeout = ctx.has_fin_timeout;
5107 facet->nf_flow.output_iface = ctx.nf_output_iface;
5108 facet->mirrors = ctx.mirrors;
5110 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
5111 if (subfacet->actions_len != odp_actions->size
5112 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
5113 free(subfacet->actions);
5114 subfacet->actions_len = odp_actions->size;
5115 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
5119 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5120 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5121 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5122 * since 'subfacet' was last updated.
5124 * Returns 0 if successful, otherwise a positive errno value. */
5126 subfacet_install(struct subfacet *subfacet,
5127 const struct nlattr *actions, size_t actions_len,
5128 struct dpif_flow_stats *stats,
5129 enum slow_path_reason slow)
5131 struct facet *facet = subfacet->facet;
5132 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5133 enum subfacet_path path = subfacet_want_path(slow);
5134 uint64_t slow_path_stub[128 / 8];
5135 enum dpif_flow_put_flags flags;
5138 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5140 flags |= DPIF_FP_ZERO_STATS;
5143 if (path == SF_SLOW_PATH) {
5144 compose_slow_path(ofproto, &facet->flow, slow,
5145 slow_path_stub, sizeof slow_path_stub,
5146 &actions, &actions_len);
5149 ret = dpif_flow_put(ofproto->backer->dpif, flags, subfacet->key,
5150 subfacet->key_len, actions, actions_len, stats);
5153 subfacet_reset_dp_stats(subfacet, stats);
5157 subfacet->path = path;
5163 subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats)
5165 return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len,
5166 stats, subfacet->slow);
5169 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5171 subfacet_uninstall(struct subfacet *subfacet)
5173 if (subfacet->path != SF_NOT_INSTALLED) {
5174 struct rule_dpif *rule = subfacet->facet->rule;
5175 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5176 struct dpif_flow_stats stats;
5179 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5180 subfacet->key_len, &stats);
5181 subfacet_reset_dp_stats(subfacet, &stats);
5183 subfacet_update_stats(subfacet, &stats);
5185 subfacet->path = SF_NOT_INSTALLED;
5187 ovs_assert(subfacet->dp_packet_count == 0);
5188 ovs_assert(subfacet->dp_byte_count == 0);
5192 /* Resets 'subfacet''s datapath statistics counters. This should be called
5193 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5194 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5195 * was reset in the datapath. 'stats' will be modified to include only
5196 * statistics new since 'subfacet' was last updated. */
5198 subfacet_reset_dp_stats(struct subfacet *subfacet,
5199 struct dpif_flow_stats *stats)
5202 && subfacet->dp_packet_count <= stats->n_packets
5203 && subfacet->dp_byte_count <= stats->n_bytes) {
5204 stats->n_packets -= subfacet->dp_packet_count;
5205 stats->n_bytes -= subfacet->dp_byte_count;
5208 subfacet->dp_packet_count = 0;
5209 subfacet->dp_byte_count = 0;
5212 /* Updates 'subfacet''s used time. The caller is responsible for calling
5213 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
5215 subfacet_update_time(struct subfacet *subfacet, long long int used)
5217 if (used > subfacet->used) {
5218 subfacet->used = used;
5219 facet_update_time(subfacet->facet, used);
5223 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5225 * Because of the meaning of a subfacet's counters, it only makes sense to do
5226 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5227 * represents a packet that was sent by hand or if it represents statistics
5228 * that have been cleared out of the datapath. */
5230 subfacet_update_stats(struct subfacet *subfacet,
5231 const struct dpif_flow_stats *stats)
5233 if (stats->n_packets || stats->used > subfacet->used) {
5234 struct facet *facet = subfacet->facet;
5236 subfacet_update_time(subfacet, stats->used);
5237 facet->packet_count += stats->n_packets;
5238 facet->byte_count += stats->n_bytes;
5239 facet->tcp_flags |= stats->tcp_flags;
5240 facet_push_stats(facet);
5241 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
5247 static struct rule_dpif *
5248 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
5250 struct rule_dpif *rule;
5252 rule = rule_dpif_lookup__(ofproto, flow, 0);
5257 return rule_dpif_miss_rule(ofproto, flow);
5260 static struct rule_dpif *
5261 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5264 struct cls_rule *cls_rule;
5265 struct classifier *cls;
5267 if (table_id >= N_TABLES) {
5271 cls = &ofproto->up.tables[table_id].cls;
5272 if (flow->nw_frag & FLOW_NW_FRAG_ANY
5273 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5274 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
5275 * are unavailable. */
5276 struct flow ofpc_normal_flow = *flow;
5277 ofpc_normal_flow.tp_src = htons(0);
5278 ofpc_normal_flow.tp_dst = htons(0);
5279 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
5281 cls_rule = classifier_lookup(cls, flow);
5283 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5286 static struct rule_dpif *
5287 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5289 struct ofport_dpif *port;
5291 port = get_ofp_port(ofproto, flow->in_port);
5293 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5294 return ofproto->miss_rule;
5297 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5298 return ofproto->no_packet_in_rule;
5300 return ofproto->miss_rule;
5304 complete_operation(struct rule_dpif *rule)
5306 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5308 rule_invalidate(rule);
5310 struct dpif_completion *c = xmalloc(sizeof *c);
5311 c->op = rule->up.pending;
5312 list_push_back(&ofproto->completions, &c->list_node);
5314 ofoperation_complete(rule->up.pending, 0);
5318 static struct rule *
5321 struct rule_dpif *rule = xmalloc(sizeof *rule);
5326 rule_dealloc(struct rule *rule_)
5328 struct rule_dpif *rule = rule_dpif_cast(rule_);
5333 rule_construct(struct rule *rule_)
5335 struct rule_dpif *rule = rule_dpif_cast(rule_);
5336 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5337 struct rule_dpif *victim;
5340 rule->packet_count = 0;
5341 rule->byte_count = 0;
5343 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5344 if (victim && !list_is_empty(&victim->facets)) {
5345 struct facet *facet;
5347 rule->facets = victim->facets;
5348 list_moved(&rule->facets);
5349 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5350 /* XXX: We're only clearing our local counters here. It's possible
5351 * that quite a few packets are unaccounted for in the datapath
5352 * statistics. These will be accounted to the new rule instead of
5353 * cleared as required. This could be fixed by clearing out the
5354 * datapath statistics for this facet, but currently it doesn't
5356 facet_reset_counters(facet);
5360 /* Must avoid list_moved() in this case. */
5361 list_init(&rule->facets);
5364 table_id = rule->up.table_id;
5366 rule->tag = victim->tag;
5367 } else if (table_id == 0) {
5372 miniflow_expand(&rule->up.cr.match.flow, &flow);
5373 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5374 ofproto->tables[table_id].basis);
5377 complete_operation(rule);
5382 rule_destruct(struct rule *rule_)
5384 struct rule_dpif *rule = rule_dpif_cast(rule_);
5385 struct facet *facet, *next_facet;
5387 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5388 facet_revalidate(facet);
5391 complete_operation(rule);
5395 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5397 struct rule_dpif *rule = rule_dpif_cast(rule_);
5398 struct facet *facet;
5400 /* Start from historical data for 'rule' itself that are no longer tracked
5401 * in facets. This counts, for example, facets that have expired. */
5402 *packets = rule->packet_count;
5403 *bytes = rule->byte_count;
5405 /* Add any statistics that are tracked by facets. This includes
5406 * statistical data recently updated by ofproto_update_stats() as well as
5407 * stats for packets that were executed "by hand" via dpif_execute(). */
5408 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5409 *packets += facet->packet_count;
5410 *bytes += facet->byte_count;
5415 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5416 struct ofpbuf *packet)
5418 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5419 struct initial_vals initial_vals;
5420 struct dpif_flow_stats stats;
5421 struct action_xlate_ctx ctx;
5422 uint64_t odp_actions_stub[1024 / 8];
5423 struct ofpbuf odp_actions;
5425 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5426 rule_credit_stats(rule, &stats);
5428 initial_vals.vlan_tci = flow->vlan_tci;
5429 initial_vals.tunnel_ip_tos = flow->tunnel.ip_tos;
5430 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5431 action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals,
5432 rule, stats.tcp_flags, packet);
5433 ctx.resubmit_stats = &stats;
5434 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
5436 execute_odp_actions(ofproto, flow, odp_actions.data,
5437 odp_actions.size, packet);
5439 ofpbuf_uninit(&odp_actions);
5443 rule_execute(struct rule *rule, const struct flow *flow,
5444 struct ofpbuf *packet)
5446 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5447 ofpbuf_delete(packet);
5452 rule_modify_actions(struct rule *rule_)
5454 struct rule_dpif *rule = rule_dpif_cast(rule_);
5456 complete_operation(rule);
5459 /* Sends 'packet' out 'ofport'.
5460 * May modify 'packet'.
5461 * Returns 0 if successful, otherwise a positive errno value. */
5463 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5465 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5466 uint64_t odp_actions_stub[1024 / 8];
5467 struct ofpbuf key, odp_actions;
5468 struct odputil_keybuf keybuf;
5473 flow_extract(packet, 0, 0, NULL, OFPP_LOCAL, &flow);
5474 if (netdev_vport_is_patch(ofport->up.netdev)) {
5475 struct ofproto_dpif *peer_ofproto;
5476 struct dpif_flow_stats stats;
5477 struct ofport_dpif *peer;
5478 struct rule_dpif *rule;
5480 peer = ofport_get_peer(ofport);
5485 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5486 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5487 netdev_vport_inc_rx(peer->up.netdev, &stats);
5489 flow.in_port = peer->up.ofp_port;
5490 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5491 rule = rule_dpif_lookup(peer_ofproto, &flow);
5492 rule_dpif_execute(rule, &flow, packet);
5497 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5499 if (ofport->tnl_port) {
5500 struct dpif_flow_stats stats;
5502 odp_port = tnl_port_send(ofport->tnl_port, &flow);
5503 if (odp_port == OVSP_NONE) {
5507 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5508 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5509 odp_put_tunnel_action(&flow.tunnel, &odp_actions);
5510 odp_put_skb_mark_action(flow.skb_mark, &odp_actions);
5512 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
5514 if (odp_port != ofport->odp_port) {
5515 eth_pop_vlan(packet);
5516 flow.vlan_tci = htons(0);
5520 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5521 odp_flow_key_from_flow(&key, &flow,
5522 ofp_port_to_odp_port(ofproto, flow.in_port));
5524 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
5526 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
5527 error = dpif_execute(ofproto->backer->dpif,
5529 odp_actions.data, odp_actions.size,
5531 ofpbuf_uninit(&odp_actions);
5534 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
5535 ofproto->up.name, odp_port, strerror(error));
5537 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
5541 /* OpenFlow to datapath action translation. */
5543 static bool may_receive(const struct ofport_dpif *, struct action_xlate_ctx *);
5544 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5545 struct action_xlate_ctx *);
5546 static void xlate_normal(struct action_xlate_ctx *);
5548 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5549 * The action will state 'slow' as the reason that the action is in the slow
5550 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5551 * dump-flows" output to see why a flow is in the slow path.)
5553 * The 'stub_size' bytes in 'stub' will be used to store the action.
5554 * 'stub_size' must be large enough for the action.
5556 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5559 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5560 enum slow_path_reason slow,
5561 uint64_t *stub, size_t stub_size,
5562 const struct nlattr **actionsp, size_t *actions_lenp)
5564 union user_action_cookie cookie;
5567 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5568 cookie.slow_path.unused = 0;
5569 cookie.slow_path.reason = slow;
5571 ofpbuf_use_stack(&buf, stub, stub_size);
5572 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
5573 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5574 odp_put_userspace_action(pid, &cookie, sizeof cookie, &buf);
5576 put_userspace_action(ofproto, &buf, flow, &cookie);
5578 *actionsp = buf.data;
5579 *actions_lenp = buf.size;
5583 put_userspace_action(const struct ofproto_dpif *ofproto,
5584 struct ofpbuf *odp_actions,
5585 const struct flow *flow,
5586 const union user_action_cookie *cookie)
5590 pid = dpif_port_get_pid(ofproto->backer->dpif,
5591 ofp_port_to_odp_port(ofproto, flow->in_port));
5593 return odp_put_userspace_action(pid, cookie, sizeof *cookie, odp_actions);
5597 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5598 ovs_be16 vlan_tci, uint32_t odp_port,
5599 unsigned int n_outputs, union user_action_cookie *cookie)
5603 cookie->type = USER_ACTION_COOKIE_SFLOW;
5604 cookie->sflow.vlan_tci = vlan_tci;
5606 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5607 * port information") for the interpretation of cookie->output. */
5608 switch (n_outputs) {
5610 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5611 cookie->sflow.output = 0x40000000 | 256;
5615 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5617 cookie->sflow.output = ifindex;
5622 /* 0x80000000 means "multiple output ports. */
5623 cookie->sflow.output = 0x80000000 | n_outputs;
5628 /* Compose SAMPLE action for sFlow. */
5630 compose_sflow_action(const struct ofproto_dpif *ofproto,
5631 struct ofpbuf *odp_actions,
5632 const struct flow *flow,
5635 uint32_t probability;
5636 union user_action_cookie cookie;
5637 size_t sample_offset, actions_offset;
5640 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5644 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5646 /* Number of packets out of UINT_MAX to sample. */
5647 probability = dpif_sflow_get_probability(ofproto->sflow);
5648 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5650 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5651 compose_sflow_cookie(ofproto, htons(0), odp_port,
5652 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5653 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
5655 nl_msg_end_nested(odp_actions, actions_offset);
5656 nl_msg_end_nested(odp_actions, sample_offset);
5657 return cookie_offset;
5660 /* SAMPLE action must be first action in any given list of actions.
5661 * At this point we do not have all information required to build it. So try to
5662 * build sample action as complete as possible. */
5664 add_sflow_action(struct action_xlate_ctx *ctx)
5666 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
5668 &ctx->flow, OVSP_NONE);
5669 ctx->sflow_odp_port = 0;
5670 ctx->sflow_n_outputs = 0;
5673 /* Fix SAMPLE action according to data collected while composing ODP actions.
5674 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
5675 * USERSPACE action's user-cookie which is required for sflow. */
5677 fix_sflow_action(struct action_xlate_ctx *ctx)
5679 const struct flow *base = &ctx->base_flow;
5680 union user_action_cookie *cookie;
5682 if (!ctx->user_cookie_offset) {
5686 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
5688 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
5690 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
5691 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
5695 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
5698 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
5699 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
5700 ovs_be64 flow_tun_id = ctx->flow.tunnel.tun_id;
5701 uint8_t flow_nw_tos = ctx->flow.nw_tos;
5702 struct priority_to_dscp *pdscp;
5703 uint32_t out_port, odp_port;
5705 /* If 'struct flow' gets additional metadata, we'll need to zero it out
5706 * before traversing a patch port. */
5707 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 19);
5710 xlate_report(ctx, "Nonexistent output port");
5712 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
5713 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
5715 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
5716 xlate_report(ctx, "STP not in forwarding state, skipping output");
5720 if (netdev_vport_is_patch(ofport->up.netdev)) {
5721 struct ofport_dpif *peer = ofport_get_peer(ofport);
5722 struct flow old_flow = ctx->flow;
5723 const struct ofproto_dpif *peer_ofproto;
5724 enum slow_path_reason special;
5725 struct ofport_dpif *in_port;
5728 xlate_report(ctx, "Nonexistent patch port peer");
5732 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5733 if (peer_ofproto->backer != ctx->ofproto->backer) {
5734 xlate_report(ctx, "Patch port peer on a different datapath");
5738 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
5739 ctx->flow.in_port = peer->up.ofp_port;
5740 ctx->flow.metadata = htonll(0);
5741 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
5742 memset(ctx->flow.regs, 0, sizeof ctx->flow.regs);
5744 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5745 special = process_special(ctx->ofproto, &ctx->flow, in_port,
5748 ctx->slow |= special;
5749 } else if (!in_port || may_receive(in_port, ctx)) {
5750 if (!in_port || stp_forward_in_state(in_port->stp_state)) {
5751 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
5753 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
5754 * learning action look at the packet, then drop it. */
5755 struct flow old_base_flow = ctx->base_flow;
5756 size_t old_size = ctx->odp_actions->size;
5757 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
5758 ctx->base_flow = old_base_flow;
5759 ctx->odp_actions->size = old_size;
5763 ctx->flow = old_flow;
5764 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5766 if (ctx->resubmit_stats) {
5767 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
5768 netdev_vport_inc_rx(peer->up.netdev, ctx->resubmit_stats);
5774 pdscp = get_priority(ofport, ctx->flow.skb_priority);
5776 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5777 ctx->flow.nw_tos |= pdscp->dscp;
5780 odp_port = ofp_port_to_odp_port(ctx->ofproto, ofp_port);
5781 if (ofport->tnl_port) {
5782 odp_port = tnl_port_send(ofport->tnl_port, &ctx->flow);
5783 if (odp_port == OVSP_NONE) {
5784 xlate_report(ctx, "Tunneling decided against output");
5788 if (ctx->resubmit_stats) {
5789 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
5791 out_port = odp_port;
5792 commit_odp_tunnel_action(&ctx->flow, &ctx->base_flow,
5795 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
5796 ctx->flow.vlan_tci);
5797 if (out_port != odp_port) {
5798 ctx->flow.vlan_tci = htons(0);
5801 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
5802 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
5804 ctx->sflow_odp_port = odp_port;
5805 ctx->sflow_n_outputs++;
5806 ctx->nf_output_iface = ofp_port;
5807 ctx->flow.tunnel.tun_id = flow_tun_id;
5808 ctx->flow.vlan_tci = flow_vlan_tci;
5809 ctx->flow.nw_tos = flow_nw_tos;
5813 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
5815 compose_output_action__(ctx, ofp_port, true);
5819 xlate_table_action(struct action_xlate_ctx *ctx,
5820 uint16_t in_port, uint8_t table_id, bool may_packet_in)
5822 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
5823 struct ofproto_dpif *ofproto = ctx->ofproto;
5824 struct rule_dpif *rule;
5825 uint16_t old_in_port;
5826 uint8_t old_table_id;
5828 old_table_id = ctx->table_id;
5829 ctx->table_id = table_id;
5831 /* Look up a flow with 'in_port' as the input port. */
5832 old_in_port = ctx->flow.in_port;
5833 ctx->flow.in_port = in_port;
5834 rule = rule_dpif_lookup__(ofproto, &ctx->flow, table_id);
5837 if (table_id > 0 && table_id < N_TABLES) {
5838 struct table_dpif *table = &ofproto->tables[table_id];
5839 if (table->other_table) {
5840 ctx->tags |= (rule && rule->tag
5842 : rule_calculate_tag(&ctx->flow,
5843 &table->other_table->mask,
5848 /* Restore the original input port. Otherwise OFPP_NORMAL and
5849 * OFPP_IN_PORT will have surprising behavior. */
5850 ctx->flow.in_port = old_in_port;
5852 if (ctx->resubmit_hook) {
5853 ctx->resubmit_hook(ctx, rule);
5856 if (rule == NULL && may_packet_in) {
5858 * check if table configuration flags
5859 * OFPTC_TABLE_MISS_CONTROLLER, default.
5860 * OFPTC_TABLE_MISS_CONTINUE,
5861 * OFPTC_TABLE_MISS_DROP
5862 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
5864 rule = rule_dpif_miss_rule(ofproto, &ctx->flow);
5868 struct rule_dpif *old_rule = ctx->rule;
5870 if (ctx->resubmit_stats) {
5871 rule_credit_stats(rule, ctx->resubmit_stats);
5876 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
5877 ctx->rule = old_rule;
5881 ctx->table_id = old_table_id;
5883 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5885 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
5886 MAX_RESUBMIT_RECURSION);
5887 ctx->max_resubmit_trigger = true;
5892 xlate_ofpact_resubmit(struct action_xlate_ctx *ctx,
5893 const struct ofpact_resubmit *resubmit)
5898 in_port = resubmit->in_port;
5899 if (in_port == OFPP_IN_PORT) {
5900 in_port = ctx->flow.in_port;
5903 table_id = resubmit->table_id;
5904 if (table_id == 255) {
5905 table_id = ctx->table_id;
5908 xlate_table_action(ctx, in_port, table_id, false);
5912 flood_packets(struct action_xlate_ctx *ctx, bool all)
5914 struct ofport_dpif *ofport;
5916 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
5917 uint16_t ofp_port = ofport->up.ofp_port;
5919 if (ofp_port == ctx->flow.in_port) {
5924 compose_output_action__(ctx, ofp_port, false);
5925 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
5926 compose_output_action(ctx, ofp_port);
5930 ctx->nf_output_iface = NF_OUT_FLOOD;
5934 execute_controller_action(struct action_xlate_ctx *ctx, int len,
5935 enum ofp_packet_in_reason reason,
5936 uint16_t controller_id)
5938 struct ofputil_packet_in pin;
5939 struct ofpbuf *packet;
5941 ctx->slow |= SLOW_CONTROLLER;
5946 packet = ofpbuf_clone(ctx->packet);
5948 if (packet->l2 && packet->l3) {
5949 struct eth_header *eh;
5950 uint16_t mpls_depth;
5952 eth_pop_vlan(packet);
5955 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
5956 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
5958 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
5959 eth_push_vlan(packet, ctx->flow.vlan_tci);
5962 mpls_depth = eth_mpls_depth(packet);
5964 if (mpls_depth < ctx->flow.mpls_depth) {
5965 push_mpls(packet, ctx->flow.dl_type, ctx->flow.mpls_lse);
5966 } else if (mpls_depth > ctx->flow.mpls_depth) {
5967 pop_mpls(packet, ctx->flow.dl_type);
5968 } else if (mpls_depth) {
5969 set_mpls_lse(packet, ctx->flow.mpls_lse);
5973 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
5974 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
5975 ctx->flow.nw_tos, ctx->flow.nw_ttl);
5979 if (ctx->flow.nw_proto == IPPROTO_TCP) {
5980 packet_set_tcp_port(packet, ctx->flow.tp_src,
5982 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
5983 packet_set_udp_port(packet, ctx->flow.tp_src,
5990 pin.packet = packet->data;
5991 pin.packet_len = packet->size;
5992 pin.reason = reason;
5993 pin.controller_id = controller_id;
5994 pin.table_id = ctx->table_id;
5995 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
5998 flow_get_metadata(&ctx->flow, &pin.fmd);
6000 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
6001 ofpbuf_delete(packet);
6005 execute_mpls_push_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type)
6007 ovs_assert(eth_type_mpls(eth_type));
6009 if (ctx->base_flow.mpls_depth) {
6010 ctx->flow.mpls_lse &= ~htonl(MPLS_BOS_MASK);
6011 ctx->flow.mpls_depth++;
6016 if (ctx->flow.dl_type == htons(ETH_TYPE_IPV6)) {
6017 label = htonl(0x2); /* IPV6 Explicit Null. */
6019 label = htonl(0x0); /* IPV4 Explicit Null. */
6021 tc = (ctx->flow.nw_tos & IP_DSCP_MASK) >> 2;
6022 ttl = ctx->flow.nw_ttl ? ctx->flow.nw_ttl : 0x40;
6023 ctx->flow.mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
6024 ctx->flow.encap_dl_type = ctx->flow.dl_type;
6025 ctx->flow.mpls_depth = 1;
6027 ctx->flow.dl_type = eth_type;
6031 execute_mpls_pop_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type)
6033 ovs_assert(eth_type_mpls(ctx->flow.dl_type));
6034 ovs_assert(!eth_type_mpls(eth_type));
6036 if (ctx->flow.mpls_depth) {
6037 ctx->flow.mpls_depth--;
6038 ctx->flow.mpls_lse = htonl(0);
6039 if (!ctx->flow.mpls_depth) {
6040 ctx->flow.dl_type = eth_type;
6041 ctx->flow.encap_dl_type = htons(0);
6047 compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
6049 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
6050 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
6054 if (ctx->flow.nw_ttl > 1) {
6060 for (i = 0; i < ids->n_controllers; i++) {
6061 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
6065 /* Stop processing for current table. */
6071 execute_set_mpls_ttl_action(struct action_xlate_ctx *ctx, uint8_t ttl)
6073 if (!eth_type_mpls(ctx->flow.dl_type)) {
6077 set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl);
6082 execute_dec_mpls_ttl_action(struct action_xlate_ctx *ctx)
6084 uint8_t ttl = mpls_lse_to_ttl(ctx->flow.mpls_lse);
6086 if (!eth_type_mpls(ctx->flow.dl_type)) {
6092 set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl);
6095 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
6097 /* Stop processing for current table. */
6103 xlate_output_action(struct action_xlate_ctx *ctx,
6104 uint16_t port, uint16_t max_len, bool may_packet_in)
6106 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
6108 ctx->nf_output_iface = NF_OUT_DROP;
6112 compose_output_action(ctx, ctx->flow.in_port);
6115 xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in);
6121 flood_packets(ctx, false);
6124 flood_packets(ctx, true);
6126 case OFPP_CONTROLLER:
6127 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
6133 if (port != ctx->flow.in_port) {
6134 compose_output_action(ctx, port);
6136 xlate_report(ctx, "skipping output to input port");
6141 if (prev_nf_output_iface == NF_OUT_FLOOD) {
6142 ctx->nf_output_iface = NF_OUT_FLOOD;
6143 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
6144 ctx->nf_output_iface = prev_nf_output_iface;
6145 } else if (prev_nf_output_iface != NF_OUT_DROP &&
6146 ctx->nf_output_iface != NF_OUT_FLOOD) {
6147 ctx->nf_output_iface = NF_OUT_MULTI;
6152 xlate_output_reg_action(struct action_xlate_ctx *ctx,
6153 const struct ofpact_output_reg *or)
6155 uint64_t port = mf_get_subfield(&or->src, &ctx->flow);
6156 if (port <= UINT16_MAX) {
6157 xlate_output_action(ctx, port, or->max_len, false);
6162 xlate_enqueue_action(struct action_xlate_ctx *ctx,
6163 const struct ofpact_enqueue *enqueue)
6165 uint16_t ofp_port = enqueue->port;
6166 uint32_t queue_id = enqueue->queue;
6167 uint32_t flow_priority, priority;
6170 /* Translate queue to priority. */
6171 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6172 queue_id, &priority);
6174 /* Fall back to ordinary output action. */
6175 xlate_output_action(ctx, enqueue->port, 0, false);
6179 /* Check output port. */
6180 if (ofp_port == OFPP_IN_PORT) {
6181 ofp_port = ctx->flow.in_port;
6182 } else if (ofp_port == ctx->flow.in_port) {
6186 /* Add datapath actions. */
6187 flow_priority = ctx->flow.skb_priority;
6188 ctx->flow.skb_priority = priority;
6189 compose_output_action(ctx, ofp_port);
6190 ctx->flow.skb_priority = flow_priority;
6192 /* Update NetFlow output port. */
6193 if (ctx->nf_output_iface == NF_OUT_DROP) {
6194 ctx->nf_output_iface = ofp_port;
6195 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
6196 ctx->nf_output_iface = NF_OUT_MULTI;
6201 xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id)
6203 uint32_t skb_priority;
6205 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6206 queue_id, &skb_priority)) {
6207 ctx->flow.skb_priority = skb_priority;
6209 /* Couldn't translate queue to a priority. Nothing to do. A warning
6210 * has already been logged. */
6214 struct xlate_reg_state {
6220 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
6222 struct ofproto_dpif *ofproto = ofproto_;
6223 struct ofport_dpif *port;
6233 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
6236 port = get_ofp_port(ofproto, ofp_port);
6237 return port ? port->may_enable : false;
6242 xlate_bundle_action(struct action_xlate_ctx *ctx,
6243 const struct ofpact_bundle *bundle)
6247 port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto);
6248 if (bundle->dst.field) {
6249 nxm_reg_load(&bundle->dst, port, &ctx->flow);
6251 xlate_output_action(ctx, port, 0, false);
6256 xlate_learn_action(struct action_xlate_ctx *ctx,
6257 const struct ofpact_learn *learn)
6259 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
6260 struct ofputil_flow_mod fm;
6261 uint64_t ofpacts_stub[1024 / 8];
6262 struct ofpbuf ofpacts;
6265 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
6266 learn_execute(learn, &ctx->flow, &fm, &ofpacts);
6268 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
6269 if (error && !VLOG_DROP_WARN(&rl)) {
6270 VLOG_WARN("learning action failed to modify flow table (%s)",
6271 ofperr_get_name(error));
6274 ofpbuf_uninit(&ofpacts);
6277 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
6278 * means "infinite". */
6280 reduce_timeout(uint16_t max, uint16_t *timeout)
6282 if (max && (!*timeout || *timeout > max)) {
6288 xlate_fin_timeout(struct action_xlate_ctx *ctx,
6289 const struct ofpact_fin_timeout *oft)
6291 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
6292 struct rule_dpif *rule = ctx->rule;
6294 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
6295 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
6300 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
6302 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
6303 ? OFPUTIL_PC_NO_RECV_STP
6304 : OFPUTIL_PC_NO_RECV)) {
6308 /* Only drop packets here if both forwarding and learning are
6309 * disabled. If just learning is enabled, we need to have
6310 * OFPP_NORMAL and the learning action have a look at the packet
6311 * before we can drop it. */
6312 if (!stp_forward_in_state(port->stp_state)
6313 && !stp_learn_in_state(port->stp_state)) {
6321 tunnel_ecn_ok(struct action_xlate_ctx *ctx)
6323 if (is_ip_any(&ctx->base_flow)
6324 && (ctx->base_flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
6325 if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
6326 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
6327 " but is not ECN capable");
6330 /* Set the ECN CE value in the tunneled packet. */
6331 ctx->flow.nw_tos |= IP_ECN_CE;
6339 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
6340 struct action_xlate_ctx *ctx)
6342 bool was_evictable = true;
6343 const struct ofpact *a;
6346 /* Don't let the rule we're working on get evicted underneath us. */
6347 was_evictable = ctx->rule->up.evictable;
6348 ctx->rule->up.evictable = false;
6350 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
6351 struct ofpact_controller *controller;
6352 const struct ofpact_metadata *metadata;
6360 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
6361 ofpact_get_OUTPUT(a)->max_len, true);
6364 case OFPACT_CONTROLLER:
6365 controller = ofpact_get_CONTROLLER(a);
6366 execute_controller_action(ctx, controller->max_len,
6368 controller->controller_id);
6371 case OFPACT_ENQUEUE:
6372 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
6375 case OFPACT_SET_VLAN_VID:
6376 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
6377 ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
6381 case OFPACT_SET_VLAN_PCP:
6382 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
6383 ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
6388 case OFPACT_STRIP_VLAN:
6389 ctx->flow.vlan_tci = htons(0);
6392 case OFPACT_PUSH_VLAN:
6393 /* XXX 802.1AD(QinQ) */
6394 ctx->flow.vlan_tci = htons(VLAN_CFI);
6397 case OFPACT_SET_ETH_SRC:
6398 memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
6402 case OFPACT_SET_ETH_DST:
6403 memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
6407 case OFPACT_SET_IPV4_SRC:
6408 ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
6411 case OFPACT_SET_IPV4_DST:
6412 ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
6415 case OFPACT_SET_IPV4_DSCP:
6416 /* OpenFlow 1.0 only supports IPv4. */
6417 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6418 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
6419 ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
6423 case OFPACT_SET_L4_SRC_PORT:
6424 ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
6427 case OFPACT_SET_L4_DST_PORT:
6428 ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
6431 case OFPACT_RESUBMIT:
6432 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
6435 case OFPACT_SET_TUNNEL:
6436 ctx->flow.tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
6439 case OFPACT_SET_QUEUE:
6440 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
6443 case OFPACT_POP_QUEUE:
6444 ctx->flow.skb_priority = ctx->orig_skb_priority;
6447 case OFPACT_REG_MOVE:
6448 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow);
6451 case OFPACT_REG_LOAD:
6452 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow);
6455 case OFPACT_STACK_PUSH:
6456 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), &ctx->flow,
6460 case OFPACT_STACK_POP:
6461 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), &ctx->flow,
6465 case OFPACT_PUSH_MPLS:
6466 execute_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
6469 case OFPACT_POP_MPLS:
6470 execute_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
6473 case OFPACT_SET_MPLS_TTL:
6474 if (execute_set_mpls_ttl_action(ctx, ofpact_get_SET_MPLS_TTL(a)->ttl)) {
6479 case OFPACT_DEC_MPLS_TTL:
6480 if (execute_dec_mpls_ttl_action(ctx)) {
6485 case OFPACT_DEC_TTL:
6486 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6492 /* Nothing to do. */
6495 case OFPACT_MULTIPATH:
6496 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow);
6500 ctx->ofproto->has_bundle_action = true;
6501 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6504 case OFPACT_OUTPUT_REG:
6505 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6509 ctx->has_learn = true;
6510 if (ctx->may_learn) {
6511 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6519 case OFPACT_FIN_TIMEOUT:
6520 ctx->has_fin_timeout = true;
6521 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6524 case OFPACT_CLEAR_ACTIONS:
6526 * Nothing to do because writa-actions is not supported for now.
6527 * When writa-actions is supported, clear-actions also must
6528 * be supported at the same time.
6532 case OFPACT_WRITE_METADATA:
6533 metadata = ofpact_get_WRITE_METADATA(a);
6534 ctx->flow.metadata &= ~metadata->mask;
6535 ctx->flow.metadata |= metadata->metadata & metadata->mask;
6538 case OFPACT_GOTO_TABLE: {
6539 /* XXX remove recursion */
6540 /* It is assumed that goto-table is last action */
6541 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6542 ovs_assert(ctx->table_id < ogt->table_id);
6543 xlate_table_action(ctx, ctx->flow.in_port, ogt->table_id, true);
6551 ctx->rule->up.evictable = was_evictable;
6556 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
6557 struct ofproto_dpif *ofproto, const struct flow *flow,
6558 const struct initial_vals *initial_vals,
6559 struct rule_dpif *rule,
6560 uint8_t tcp_flags, const struct ofpbuf *packet)
6562 ovs_be64 initial_tun_id = flow->tunnel.tun_id;
6564 /* Flow initialization rules:
6565 * - 'base_flow' must match the kernel's view of the packet at the
6566 * time that action processing starts. 'flow' represents any
6567 * transformations we wish to make through actions.
6568 * - By default 'base_flow' and 'flow' are the same since the input
6569 * packet matches the output before any actions are applied.
6570 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
6571 * of the received packet as seen by the kernel. If we later output
6572 * to another device without any modifications this will cause us to
6573 * insert a new tag since the original one was stripped off by the
6575 * - Tunnel 'flow' is largely cleared when transitioning between
6576 * the input and output stages since it does not make sense to output
6577 * a packet with the exact headers that it was received with (i.e.
6578 * the destination IP is us). The one exception is the tun_id, which
6579 * is preserved to allow use in later resubmit lookups and loads into
6581 * - Tunnel 'base_flow' is completely cleared since that is what the
6582 * kernel does. If we wish to maintain the original values an action
6583 * needs to be generated. */
6585 ctx->ofproto = ofproto;
6587 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
6588 ctx->base_flow = ctx->flow;
6589 ctx->base_flow.vlan_tci = initial_vals->vlan_tci;
6590 ctx->base_flow.tunnel.ip_tos = initial_vals->tunnel_ip_tos;
6591 ctx->flow.tunnel.tun_id = initial_tun_id;
6593 ctx->packet = packet;
6594 ctx->may_learn = packet != NULL;
6595 ctx->tcp_flags = tcp_flags;
6596 ctx->resubmit_hook = NULL;
6597 ctx->report_hook = NULL;
6598 ctx->resubmit_stats = NULL;
6601 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
6602 * into datapath actions in 'odp_actions', using 'ctx'. */
6604 xlate_actions(struct action_xlate_ctx *ctx,
6605 const struct ofpact *ofpacts, size_t ofpacts_len,
6606 struct ofpbuf *odp_actions)
6608 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
6609 * that in the future we always keep a copy of the original flow for
6610 * tracing purposes. */
6611 static bool hit_resubmit_limit;
6613 enum slow_path_reason special;
6614 struct ofport_dpif *in_port;
6615 struct flow orig_flow;
6617 COVERAGE_INC(ofproto_dpif_xlate);
6619 ofpbuf_clear(odp_actions);
6620 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
6622 ctx->odp_actions = odp_actions;
6625 ctx->has_learn = false;
6626 ctx->has_normal = false;
6627 ctx->has_fin_timeout = false;
6628 ctx->nf_output_iface = NF_OUT_DROP;
6631 ctx->max_resubmit_trigger = false;
6632 ctx->orig_skb_priority = ctx->flow.skb_priority;
6636 ofpbuf_use_stub(&ctx->stack, ctx->init_stack, sizeof ctx->init_stack);
6638 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
6639 /* Do this conditionally because the copy is expensive enough that it
6640 * shows up in profiles. */
6641 orig_flow = ctx->flow;
6644 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
6645 switch (ctx->ofproto->up.frag_handling) {
6646 case OFPC_FRAG_NORMAL:
6647 /* We must pretend that transport ports are unavailable. */
6648 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
6649 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
6652 case OFPC_FRAG_DROP:
6655 case OFPC_FRAG_REASM:
6658 case OFPC_FRAG_NX_MATCH:
6659 /* Nothing to do. */
6662 case OFPC_INVALID_TTL_TO_CONTROLLER:
6667 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
6668 special = process_special(ctx->ofproto, &ctx->flow, in_port, ctx->packet);
6670 ctx->slow |= special;
6672 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6673 struct initial_vals initial_vals;
6674 uint32_t local_odp_port;
6676 initial_vals.vlan_tci = ctx->base_flow.vlan_tci;
6677 initial_vals.tunnel_ip_tos = ctx->base_flow.tunnel.ip_tos;
6679 add_sflow_action(ctx);
6681 if (tunnel_ecn_ok(ctx) && (!in_port || may_receive(in_port, ctx))) {
6682 do_xlate_actions(ofpacts, ofpacts_len, ctx);
6684 /* We've let OFPP_NORMAL and the learning action look at the
6685 * packet, so drop it now if forwarding is disabled. */
6686 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
6687 ofpbuf_clear(ctx->odp_actions);
6688 add_sflow_action(ctx);
6692 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
6693 if (!hit_resubmit_limit) {
6694 /* We didn't record the original flow. Make sure we do from
6696 hit_resubmit_limit = true;
6697 } else if (!VLOG_DROP_ERR(&trace_rl)) {
6698 struct ds ds = DS_EMPTY_INITIALIZER;
6700 ofproto_trace(ctx->ofproto, &orig_flow, ctx->packet,
6701 &initial_vals, &ds);
6702 VLOG_ERR("Trace triggered by excessive resubmit "
6703 "recursion:\n%s", ds_cstr(&ds));
6708 local_odp_port = ofp_port_to_odp_port(ctx->ofproto, OFPP_LOCAL);
6709 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
6711 ctx->odp_actions->data,
6712 ctx->odp_actions->size)) {
6713 ctx->slow |= SLOW_IN_BAND;
6715 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
6717 compose_output_action(ctx, OFPP_LOCAL);
6720 if (ctx->ofproto->has_mirrors) {
6721 add_mirror_actions(ctx, &orig_flow);
6723 fix_sflow_action(ctx);
6726 ofpbuf_uninit(&ctx->stack);
6729 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
6730 * into datapath actions, using 'ctx', and discards the datapath actions. */
6732 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
6733 const struct ofpact *ofpacts,
6736 uint64_t odp_actions_stub[1024 / 8];
6737 struct ofpbuf odp_actions;
6739 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
6740 xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions);
6741 ofpbuf_uninit(&odp_actions);
6745 xlate_report(struct action_xlate_ctx *ctx, const char *s)
6747 if (ctx->report_hook) {
6748 ctx->report_hook(ctx, s);
6752 /* OFPP_NORMAL implementation. */
6754 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
6756 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
6757 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
6758 * the bundle on which the packet was received, returns the VLAN to which the
6761 * Both 'vid' and the return value are in the range 0...4095. */
6763 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
6765 switch (in_bundle->vlan_mode) {
6766 case PORT_VLAN_ACCESS:
6767 return in_bundle->vlan;
6770 case PORT_VLAN_TRUNK:
6773 case PORT_VLAN_NATIVE_UNTAGGED:
6774 case PORT_VLAN_NATIVE_TAGGED:
6775 return vid ? vid : in_bundle->vlan;
6782 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
6783 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
6786 * 'vid' should be the VID obtained from the 802.1Q header that was received as
6787 * part of a packet (specify 0 if there was no 802.1Q header), in the range
6790 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
6792 /* Allow any VID on the OFPP_NONE port. */
6793 if (in_bundle == &ofpp_none_bundle) {
6797 switch (in_bundle->vlan_mode) {
6798 case PORT_VLAN_ACCESS:
6801 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6802 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
6803 "packet received on port %s configured as VLAN "
6804 "%"PRIu16" access port",
6805 in_bundle->ofproto->up.name, vid,
6806 in_bundle->name, in_bundle->vlan);
6812 case PORT_VLAN_NATIVE_UNTAGGED:
6813 case PORT_VLAN_NATIVE_TAGGED:
6815 /* Port must always carry its native VLAN. */
6819 case PORT_VLAN_TRUNK:
6820 if (!ofbundle_includes_vlan(in_bundle, vid)) {
6822 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6823 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
6824 "received on port %s not configured for trunking "
6826 in_bundle->ofproto->up.name, vid,
6827 in_bundle->name, vid);
6839 /* Given 'vlan', the VLAN that a packet belongs to, and
6840 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
6841 * that should be included in the 802.1Q header. (If the return value is 0,
6842 * then the 802.1Q header should only be included in the packet if there is a
6845 * Both 'vlan' and the return value are in the range 0...4095. */
6847 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
6849 switch (out_bundle->vlan_mode) {
6850 case PORT_VLAN_ACCESS:
6853 case PORT_VLAN_TRUNK:
6854 case PORT_VLAN_NATIVE_TAGGED:
6857 case PORT_VLAN_NATIVE_UNTAGGED:
6858 return vlan == out_bundle->vlan ? 0 : vlan;
6866 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
6869 struct ofport_dpif *port;
6871 ovs_be16 tci, old_tci;
6873 vid = output_vlan_to_vid(out_bundle, vlan);
6874 if (!out_bundle->bond) {
6875 port = ofbundle_get_a_port(out_bundle);
6877 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
6880 /* No slaves enabled, so drop packet. */
6885 old_tci = ctx->flow.vlan_tci;
6887 if (tci || out_bundle->use_priority_tags) {
6888 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
6890 tci |= htons(VLAN_CFI);
6893 ctx->flow.vlan_tci = tci;
6895 compose_output_action(ctx, port->up.ofp_port);
6896 ctx->flow.vlan_tci = old_tci;
6900 mirror_mask_ffs(mirror_mask_t mask)
6902 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
6907 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
6909 return (bundle->vlan_mode != PORT_VLAN_ACCESS
6910 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
6914 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
6916 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
6919 /* Returns an arbitrary interface within 'bundle'. */
6920 static struct ofport_dpif *
6921 ofbundle_get_a_port(const struct ofbundle *bundle)
6923 return CONTAINER_OF(list_front(&bundle->ports),
6924 struct ofport_dpif, bundle_node);
6928 vlan_is_mirrored(const struct ofmirror *m, int vlan)
6930 return !m->vlans || bitmap_is_set(m->vlans, vlan);
6934 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
6936 struct ofproto_dpif *ofproto = ctx->ofproto;
6937 mirror_mask_t mirrors;
6938 struct ofbundle *in_bundle;
6941 const struct nlattr *a;
6944 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
6945 ctx->packet != NULL, NULL);
6949 mirrors = in_bundle->src_mirrors;
6951 /* Drop frames on bundles reserved for mirroring. */
6952 if (in_bundle->mirror_out) {
6953 if (ctx->packet != NULL) {
6954 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6955 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
6956 "%s, which is reserved exclusively for mirroring",
6957 ctx->ofproto->up.name, in_bundle->name);
6963 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
6964 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
6967 vlan = input_vid_to_vlan(in_bundle, vid);
6969 /* Look at the output ports to check for destination selections. */
6971 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
6972 ctx->odp_actions->size) {
6973 enum ovs_action_attr type = nl_attr_type(a);
6974 struct ofport_dpif *ofport;
6976 if (type != OVS_ACTION_ATTR_OUTPUT) {
6980 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
6981 if (ofport && ofport->bundle) {
6982 mirrors |= ofport->bundle->dst_mirrors;
6990 /* Restore the original packet before adding the mirror actions. */
6991 ctx->flow = *orig_flow;
6996 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
6998 if (!vlan_is_mirrored(m, vlan)) {
6999 mirrors = zero_rightmost_1bit(mirrors);
7003 mirrors &= ~m->dup_mirrors;
7004 ctx->mirrors |= m->dup_mirrors;
7006 output_normal(ctx, m->out, vlan);
7007 } else if (vlan != m->out_vlan
7008 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
7009 struct ofbundle *bundle;
7011 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
7012 if (ofbundle_includes_vlan(bundle, m->out_vlan)
7013 && !bundle->mirror_out) {
7014 output_normal(ctx, bundle, m->out_vlan);
7022 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
7023 uint64_t packets, uint64_t bytes)
7029 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
7032 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7035 /* In normal circumstances 'm' will not be NULL. However,
7036 * if mirrors are reconfigured, we can temporarily get out
7037 * of sync in facet_revalidate(). We could "correct" the
7038 * mirror list before reaching here, but doing that would
7039 * not properly account the traffic stats we've currently
7040 * accumulated for previous mirror configuration. */
7044 m->packet_count += packets;
7045 m->byte_count += bytes;
7049 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
7050 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
7051 * indicate this; newer upstream kernels use gratuitous ARP requests. */
7053 is_gratuitous_arp(const struct flow *flow)
7055 return (flow->dl_type == htons(ETH_TYPE_ARP)
7056 && eth_addr_is_broadcast(flow->dl_dst)
7057 && (flow->nw_proto == ARP_OP_REPLY
7058 || (flow->nw_proto == ARP_OP_REQUEST
7059 && flow->nw_src == flow->nw_dst)));
7063 update_learning_table(struct ofproto_dpif *ofproto,
7064 const struct flow *flow, int vlan,
7065 struct ofbundle *in_bundle)
7067 struct mac_entry *mac;
7069 /* Don't learn the OFPP_NONE port. */
7070 if (in_bundle == &ofpp_none_bundle) {
7074 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
7078 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
7079 if (is_gratuitous_arp(flow)) {
7080 /* We don't want to learn from gratuitous ARP packets that are
7081 * reflected back over bond slaves so we lock the learning table. */
7082 if (!in_bundle->bond) {
7083 mac_entry_set_grat_arp_lock(mac);
7084 } else if (mac_entry_is_grat_arp_locked(mac)) {
7089 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
7090 /* The log messages here could actually be useful in debugging,
7091 * so keep the rate limit relatively high. */
7092 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
7093 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
7094 "on port %s in VLAN %d",
7095 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
7096 in_bundle->name, vlan);
7098 mac->port.p = in_bundle;
7099 tag_set_add(&ofproto->backer->revalidate_set,
7100 mac_learning_changed(ofproto->ml, mac));
7104 static struct ofbundle *
7105 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
7106 bool warn, struct ofport_dpif **in_ofportp)
7108 struct ofport_dpif *ofport;
7110 /* Find the port and bundle for the received packet. */
7111 ofport = get_ofp_port(ofproto, in_port);
7113 *in_ofportp = ofport;
7115 if (ofport && ofport->bundle) {
7116 return ofport->bundle;
7119 /* Special-case OFPP_NONE, which a controller may use as the ingress
7120 * port for traffic that it is sourcing. */
7121 if (in_port == OFPP_NONE) {
7122 return &ofpp_none_bundle;
7125 /* Odd. A few possible reasons here:
7127 * - We deleted a port but there are still a few packets queued up
7130 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
7131 * we don't know about.
7133 * - The ofproto client didn't configure the port as part of a bundle.
7134 * This is particularly likely to happen if a packet was received on the
7135 * port after it was created, but before the client had a chance to
7136 * configure its bundle.
7139 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7141 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
7142 "port %"PRIu16, ofproto->up.name, in_port);
7147 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
7148 * dropped. Returns true if they may be forwarded, false if they should be
7151 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
7152 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
7154 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
7155 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
7156 * checked by input_vid_is_valid().
7158 * May also add tags to '*tags', although the current implementation only does
7159 * so in one special case.
7162 is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port,
7165 struct ofproto_dpif *ofproto = ctx->ofproto;
7166 struct flow *flow = &ctx->flow;
7167 struct ofbundle *in_bundle = in_port->bundle;
7169 /* Drop frames for reserved multicast addresses
7170 * only if forward_bpdu option is absent. */
7171 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
7172 xlate_report(ctx, "packet has reserved destination MAC, dropping");
7176 if (in_bundle->bond) {
7177 struct mac_entry *mac;
7179 switch (bond_check_admissibility(in_bundle->bond, in_port,
7180 flow->dl_dst, &ctx->tags)) {
7185 xlate_report(ctx, "bonding refused admissibility, dropping");
7188 case BV_DROP_IF_MOVED:
7189 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
7190 if (mac && mac->port.p != in_bundle &&
7191 (!is_gratuitous_arp(flow)
7192 || mac_entry_is_grat_arp_locked(mac))) {
7193 xlate_report(ctx, "SLB bond thinks this packet looped back, "
7205 xlate_normal(struct action_xlate_ctx *ctx)
7207 struct ofport_dpif *in_port;
7208 struct ofbundle *in_bundle;
7209 struct mac_entry *mac;
7213 ctx->has_normal = true;
7215 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
7216 ctx->packet != NULL, &in_port);
7218 xlate_report(ctx, "no input bundle, dropping");
7222 /* Drop malformed frames. */
7223 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
7224 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
7225 if (ctx->packet != NULL) {
7226 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7227 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
7228 "VLAN tag received on port %s",
7229 ctx->ofproto->up.name, in_bundle->name);
7231 xlate_report(ctx, "partial VLAN tag, dropping");
7235 /* Drop frames on bundles reserved for mirroring. */
7236 if (in_bundle->mirror_out) {
7237 if (ctx->packet != NULL) {
7238 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7239 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7240 "%s, which is reserved exclusively for mirroring",
7241 ctx->ofproto->up.name, in_bundle->name);
7243 xlate_report(ctx, "input port is mirror output port, dropping");
7248 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
7249 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
7250 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
7253 vlan = input_vid_to_vlan(in_bundle, vid);
7255 /* Check other admissibility requirements. */
7256 if (in_port && !is_admissible(ctx, in_port, vlan)) {
7260 /* Learn source MAC. */
7261 if (ctx->may_learn) {
7262 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
7265 /* Determine output bundle. */
7266 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
7269 if (mac->port.p != in_bundle) {
7270 xlate_report(ctx, "forwarding to learned port");
7271 output_normal(ctx, mac->port.p, vlan);
7273 xlate_report(ctx, "learned port is input port, dropping");
7276 struct ofbundle *bundle;
7278 xlate_report(ctx, "no learned MAC for destination, flooding");
7279 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
7280 if (bundle != in_bundle
7281 && ofbundle_includes_vlan(bundle, vlan)
7282 && bundle->floodable
7283 && !bundle->mirror_out) {
7284 output_normal(ctx, bundle, vlan);
7287 ctx->nf_output_iface = NF_OUT_FLOOD;
7291 /* Optimized flow revalidation.
7293 * It's a difficult problem, in general, to tell which facets need to have
7294 * their actions recalculated whenever the OpenFlow flow table changes. We
7295 * don't try to solve that general problem: for most kinds of OpenFlow flow
7296 * table changes, we recalculate the actions for every facet. This is
7297 * relatively expensive, but it's good enough if the OpenFlow flow table
7298 * doesn't change very often.
7300 * However, we can expect one particular kind of OpenFlow flow table change to
7301 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
7302 * of CPU on revalidating every facet whenever MAC learning modifies the flow
7303 * table, we add a special case that applies to flow tables in which every rule
7304 * has the same form (that is, the same wildcards), except that the table is
7305 * also allowed to have a single "catch-all" flow that matches all packets. We
7306 * optimize this case by tagging all of the facets that resubmit into the table
7307 * and invalidating the same tag whenever a flow changes in that table. The
7308 * end result is that we revalidate just the facets that need it (and sometimes
7309 * a few more, but not all of the facets or even all of the facets that
7310 * resubmit to the table modified by MAC learning). */
7312 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
7313 * into an OpenFlow table with the given 'basis'. */
7315 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
7318 if (minimask_is_catchall(mask)) {
7321 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
7322 return tag_create_deterministic(hash);
7326 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
7327 * taggability of that table.
7329 * This function must be called after *each* change to a flow table. If you
7330 * skip calling it on some changes then the pointer comparisons at the end can
7331 * be invalid if you get unlucky. For example, if a flow removal causes a
7332 * cls_table to be destroyed and then a flow insertion causes a cls_table with
7333 * different wildcards to be created with the same address, then this function
7334 * will incorrectly skip revalidation. */
7336 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
7338 struct table_dpif *table = &ofproto->tables[table_id];
7339 const struct oftable *oftable = &ofproto->up.tables[table_id];
7340 struct cls_table *catchall, *other;
7341 struct cls_table *t;
7343 catchall = other = NULL;
7345 switch (hmap_count(&oftable->cls.tables)) {
7347 /* We could tag this OpenFlow table but it would make the logic a
7348 * little harder and it's a corner case that doesn't seem worth it
7354 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
7355 if (cls_table_is_catchall(t)) {
7357 } else if (!other) {
7360 /* Indicate that we can't tag this by setting both tables to
7361 * NULL. (We know that 'catchall' is already NULL.) */
7368 /* Can't tag this table. */
7372 if (table->catchall_table != catchall || table->other_table != other) {
7373 table->catchall_table = catchall;
7374 table->other_table = other;
7375 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7379 /* Given 'rule' that has changed in some way (either it is a rule being
7380 * inserted, a rule being deleted, or a rule whose actions are being
7381 * modified), marks facets for revalidation to ensure that packets will be
7382 * forwarded correctly according to the new state of the flow table.
7384 * This function must be called after *each* change to a flow table. See
7385 * the comment on table_update_taggable() for more information. */
7387 rule_invalidate(const struct rule_dpif *rule)
7389 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
7391 table_update_taggable(ofproto, rule->up.table_id);
7393 if (!ofproto->backer->need_revalidate) {
7394 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
7396 if (table->other_table && rule->tag) {
7397 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
7399 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7405 set_frag_handling(struct ofproto *ofproto_,
7406 enum ofp_config_flags frag_handling)
7408 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7409 if (frag_handling != OFPC_FRAG_REASM) {
7410 ofproto->backer->need_revalidate = REV_RECONFIGURE;
7418 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
7419 const struct flow *flow,
7420 const struct ofpact *ofpacts, size_t ofpacts_len)
7422 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7423 struct initial_vals initial_vals;
7424 struct odputil_keybuf keybuf;
7425 struct dpif_flow_stats stats;
7429 struct action_xlate_ctx ctx;
7430 uint64_t odp_actions_stub[1024 / 8];
7431 struct ofpbuf odp_actions;
7433 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
7434 odp_flow_key_from_flow(&key, flow,
7435 ofp_port_to_odp_port(ofproto, flow->in_port));
7437 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
7439 initial_vals.vlan_tci = flow->vlan_tci;
7440 initial_vals.tunnel_ip_tos = 0;
7441 action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals, NULL,
7442 packet_get_tcp_flags(packet, flow), packet);
7443 ctx.resubmit_stats = &stats;
7445 ofpbuf_use_stub(&odp_actions,
7446 odp_actions_stub, sizeof odp_actions_stub);
7447 xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions);
7448 dpif_execute(ofproto->backer->dpif, key.data, key.size,
7449 odp_actions.data, odp_actions.size, packet);
7450 ofpbuf_uninit(&odp_actions);
7458 set_netflow(struct ofproto *ofproto_,
7459 const struct netflow_options *netflow_options)
7461 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7463 if (netflow_options) {
7464 if (!ofproto->netflow) {
7465 ofproto->netflow = netflow_create();
7467 return netflow_set_options(ofproto->netflow, netflow_options);
7469 netflow_destroy(ofproto->netflow);
7470 ofproto->netflow = NULL;
7476 get_netflow_ids(const struct ofproto *ofproto_,
7477 uint8_t *engine_type, uint8_t *engine_id)
7479 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7481 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
7485 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
7487 if (!facet_is_controller_flow(facet) &&
7488 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
7489 struct subfacet *subfacet;
7490 struct ofexpired expired;
7492 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
7493 if (subfacet->path == SF_FAST_PATH) {
7494 struct dpif_flow_stats stats;
7496 subfacet_reinstall(subfacet, &stats);
7497 subfacet_update_stats(subfacet, &stats);
7501 expired.flow = facet->flow;
7502 expired.packet_count = facet->packet_count;
7503 expired.byte_count = facet->byte_count;
7504 expired.used = facet->used;
7505 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
7510 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
7512 struct facet *facet;
7514 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7515 send_active_timeout(ofproto, facet);
7519 static struct ofproto_dpif *
7520 ofproto_dpif_lookup(const char *name)
7522 struct ofproto_dpif *ofproto;
7524 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
7525 hash_string(name, 0), &all_ofproto_dpifs) {
7526 if (!strcmp(ofproto->up.name, name)) {
7534 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
7535 const char *argv[], void *aux OVS_UNUSED)
7537 struct ofproto_dpif *ofproto;
7540 ofproto = ofproto_dpif_lookup(argv[1]);
7542 unixctl_command_reply_error(conn, "no such bridge");
7545 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7547 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7548 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7552 unixctl_command_reply(conn, "table successfully flushed");
7556 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
7557 const char *argv[], void *aux OVS_UNUSED)
7559 struct ds ds = DS_EMPTY_INITIALIZER;
7560 const struct ofproto_dpif *ofproto;
7561 const struct mac_entry *e;
7563 ofproto = ofproto_dpif_lookup(argv[1]);
7565 unixctl_command_reply_error(conn, "no such bridge");
7569 ds_put_cstr(&ds, " port VLAN MAC Age\n");
7570 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
7571 struct ofbundle *bundle = e->port.p;
7572 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
7573 ofbundle_get_a_port(bundle)->odp_port,
7574 e->vlan, ETH_ADDR_ARGS(e->mac),
7575 mac_entry_age(ofproto->ml, e));
7577 unixctl_command_reply(conn, ds_cstr(&ds));
7582 struct action_xlate_ctx ctx;
7588 trace_format_rule(struct ds *result, uint8_t table_id, int level,
7589 const struct rule_dpif *rule)
7591 ds_put_char_multiple(result, '\t', level);
7593 ds_put_cstr(result, "No match\n");
7597 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
7598 table_id, ntohll(rule->up.flow_cookie));
7599 cls_rule_format(&rule->up.cr, result);
7600 ds_put_char(result, '\n');
7602 ds_put_char_multiple(result, '\t', level);
7603 ds_put_cstr(result, "OpenFlow ");
7604 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
7605 ds_put_char(result, '\n');
7609 trace_format_flow(struct ds *result, int level, const char *title,
7610 struct trace_ctx *trace)
7612 ds_put_char_multiple(result, '\t', level);
7613 ds_put_format(result, "%s: ", title);
7614 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
7615 ds_put_cstr(result, "unchanged");
7617 flow_format(result, &trace->ctx.flow);
7618 trace->flow = trace->ctx.flow;
7620 ds_put_char(result, '\n');
7624 trace_format_regs(struct ds *result, int level, const char *title,
7625 struct trace_ctx *trace)
7629 ds_put_char_multiple(result, '\t', level);
7630 ds_put_format(result, "%s:", title);
7631 for (i = 0; i < FLOW_N_REGS; i++) {
7632 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
7634 ds_put_char(result, '\n');
7638 trace_format_odp(struct ds *result, int level, const char *title,
7639 struct trace_ctx *trace)
7641 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
7643 ds_put_char_multiple(result, '\t', level);
7644 ds_put_format(result, "%s: ", title);
7645 format_odp_actions(result, odp_actions->data, odp_actions->size);
7646 ds_put_char(result, '\n');
7650 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
7652 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7653 struct ds *result = trace->result;
7655 ds_put_char(result, '\n');
7656 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
7657 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
7658 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
7659 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
7663 trace_report(struct action_xlate_ctx *ctx, const char *s)
7665 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7666 struct ds *result = trace->result;
7668 ds_put_char_multiple(result, '\t', ctx->recurse);
7669 ds_put_cstr(result, s);
7670 ds_put_char(result, '\n');
7674 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
7675 void *aux OVS_UNUSED)
7677 const char *dpname = argv[1];
7678 struct ofproto_dpif *ofproto;
7679 struct ofpbuf odp_key;
7680 struct ofpbuf *packet;
7681 struct initial_vals initial_vals;
7687 ofpbuf_init(&odp_key, 0);
7690 ofproto = ofproto_dpif_lookup(dpname);
7692 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
7696 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
7697 /* ofproto/trace dpname flow [-generate] */
7698 const char *flow_s = argv[2];
7699 const char *generate_s = argv[3];
7701 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
7702 * flow. We guess which type it is based on whether 'flow_s' contains
7703 * an '(', since a datapath flow always contains '(') but an
7704 * OpenFlow-like flow should not (in fact it's allowed but I believe
7705 * that's not documented anywhere).
7707 * An alternative would be to try to parse 'flow_s' both ways, but then
7708 * it would be tricky giving a sensible error message. After all, do
7709 * you just say "syntax error" or do you present both error messages?
7710 * Both choices seem lousy. */
7711 if (strchr(flow_s, '(')) {
7714 /* Convert string to datapath key. */
7715 ofpbuf_init(&odp_key, 0);
7716 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
7718 unixctl_command_reply_error(conn, "Bad flow syntax");
7722 /* XXX: Since we allow the user to specify an ofproto, it's
7723 * possible they will specify a different ofproto than the one the
7724 * port actually belongs too. Ideally we should simply remove the
7725 * ability to specify the ofproto. */
7726 if (ofproto_receive(ofproto->backer, NULL, odp_key.data,
7727 odp_key.size, &flow, NULL, NULL, NULL,
7729 unixctl_command_reply_error(conn, "Invalid flow");
7735 error_s = parse_ofp_exact_flow(&flow, argv[2]);
7737 unixctl_command_reply_error(conn, error_s);
7742 initial_vals.vlan_tci = flow.vlan_tci;
7743 initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos;
7746 /* Generate a packet, if requested. */
7748 packet = ofpbuf_new(0);
7749 flow_compose(packet, &flow);
7751 } else if (argc == 7) {
7752 /* ofproto/trace dpname priority tun_id in_port mark packet */
7753 const char *priority_s = argv[2];
7754 const char *tun_id_s = argv[3];
7755 const char *in_port_s = argv[4];
7756 const char *mark_s = argv[5];
7757 const char *packet_s = argv[6];
7758 uint32_t in_port = atoi(in_port_s);
7759 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
7760 uint32_t priority = atoi(priority_s);
7761 uint32_t mark = atoi(mark_s);
7764 msg = eth_from_hex(packet_s, &packet);
7766 unixctl_command_reply_error(conn, msg);
7770 ds_put_cstr(&result, "Packet: ");
7771 s = ofp_packet_to_string(packet->data, packet->size);
7772 ds_put_cstr(&result, s);
7775 flow_extract(packet, priority, mark, NULL, in_port, &flow);
7776 flow.tunnel.tun_id = tun_id;
7777 initial_vals.vlan_tci = flow.vlan_tci;
7778 initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos;
7780 unixctl_command_reply_error(conn, "Bad command syntax");
7784 ofproto_trace(ofproto, &flow, packet, &initial_vals, &result);
7785 unixctl_command_reply(conn, ds_cstr(&result));
7788 ds_destroy(&result);
7789 ofpbuf_delete(packet);
7790 ofpbuf_uninit(&odp_key);
7794 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
7795 const struct ofpbuf *packet,
7796 const struct initial_vals *initial_vals, struct ds *ds)
7798 struct rule_dpif *rule;
7800 ds_put_cstr(ds, "Flow: ");
7801 flow_format(ds, flow);
7802 ds_put_char(ds, '\n');
7804 rule = rule_dpif_lookup(ofproto, flow);
7806 trace_format_rule(ds, 0, 0, rule);
7807 if (rule == ofproto->miss_rule) {
7808 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
7809 } else if (rule == ofproto->no_packet_in_rule) {
7810 ds_put_cstr(ds, "\nNo match, packets dropped because "
7811 "OFPPC_NO_PACKET_IN is set on in_port.\n");
7815 uint64_t odp_actions_stub[1024 / 8];
7816 struct ofpbuf odp_actions;
7818 struct trace_ctx trace;
7821 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
7824 ofpbuf_use_stub(&odp_actions,
7825 odp_actions_stub, sizeof odp_actions_stub);
7826 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_vals,
7827 rule, tcp_flags, packet);
7828 trace.ctx.resubmit_hook = trace_resubmit;
7829 trace.ctx.report_hook = trace_report;
7830 xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len,
7833 ds_put_char(ds, '\n');
7834 trace_format_flow(ds, 0, "Final flow", &trace);
7835 ds_put_cstr(ds, "Datapath actions: ");
7836 format_odp_actions(ds, odp_actions.data, odp_actions.size);
7837 ofpbuf_uninit(&odp_actions);
7839 if (trace.ctx.slow) {
7840 enum slow_path_reason slow;
7842 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
7843 "slow path because it:");
7844 for (slow = trace.ctx.slow; slow; ) {
7845 enum slow_path_reason bit = rightmost_1bit(slow);
7849 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
7852 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
7855 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
7858 ds_put_cstr(ds, "\n\t- Needs in-band special case "
7861 ds_put_cstr(ds, "\n\t (The datapath actions are "
7862 "incomplete--for complete actions, "
7863 "please supply a packet.)");
7866 case SLOW_CONTROLLER:
7867 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
7868 "to the OpenFlow controller.");
7871 ds_put_cstr(ds, "\n\t- Needs more specific matching "
7872 "than the datapath supports.");
7879 if (slow & ~SLOW_MATCH) {
7880 ds_put_cstr(ds, "\nThe datapath actions above do not reflect "
7881 "the special slow-path processing.");
7888 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7889 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7892 unixctl_command_reply(conn, NULL);
7896 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7897 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7900 unixctl_command_reply(conn, NULL);
7903 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
7904 * 'reply' describing the results. */
7906 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
7908 struct facet *facet;
7912 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7913 if (!facet_check_consistency(facet)) {
7918 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
7922 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
7923 ofproto->up.name, errors);
7925 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
7930 ofproto_dpif_self_check(struct unixctl_conn *conn,
7931 int argc, const char *argv[], void *aux OVS_UNUSED)
7933 struct ds reply = DS_EMPTY_INITIALIZER;
7934 struct ofproto_dpif *ofproto;
7937 ofproto = ofproto_dpif_lookup(argv[1]);
7939 unixctl_command_reply_error(conn, "Unknown ofproto (use "
7940 "ofproto/list for help)");
7943 ofproto_dpif_self_check__(ofproto, &reply);
7945 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7946 ofproto_dpif_self_check__(ofproto, &reply);
7950 unixctl_command_reply(conn, ds_cstr(&reply));
7954 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
7955 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
7956 * to destroy 'ofproto_shash' and free the returned value. */
7957 static const struct shash_node **
7958 get_ofprotos(struct shash *ofproto_shash)
7960 const struct ofproto_dpif *ofproto;
7962 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7963 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
7964 shash_add_nocopy(ofproto_shash, name, ofproto);
7967 return shash_sort(ofproto_shash);
7971 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
7972 const char *argv[] OVS_UNUSED,
7973 void *aux OVS_UNUSED)
7975 struct ds ds = DS_EMPTY_INITIALIZER;
7976 struct shash ofproto_shash;
7977 const struct shash_node **sorted_ofprotos;
7980 shash_init(&ofproto_shash);
7981 sorted_ofprotos = get_ofprotos(&ofproto_shash);
7982 for (i = 0; i < shash_count(&ofproto_shash); i++) {
7983 const struct shash_node *node = sorted_ofprotos[i];
7984 ds_put_format(&ds, "%s\n", node->name);
7987 shash_destroy(&ofproto_shash);
7988 free(sorted_ofprotos);
7990 unixctl_command_reply(conn, ds_cstr(&ds));
7995 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
7997 struct dpif_dp_stats s;
7998 const struct shash_node **ports;
8001 dpif_get_dp_stats(ofproto->backer->dpif, &s);
8003 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
8004 dpif_name(ofproto->backer->dpif));
8005 /* xxx It would be better to show bridge-specific stats instead
8006 * xxx of dp ones. */
8008 "\tlookups: hit:%"PRIu64" missed:%"PRIu64" lost:%"PRIu64"\n",
8009 s.n_hit, s.n_missed, s.n_lost);
8010 ds_put_format(ds, "\tflows: %zu\n",
8011 hmap_count(&ofproto->subfacets));
8013 ports = shash_sort(&ofproto->up.port_by_name);
8014 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
8015 const struct shash_node *node = ports[i];
8016 struct ofport *ofport = node->data;
8017 const char *name = netdev_get_name(ofport->netdev);
8018 const char *type = netdev_get_type(ofport->netdev);
8021 ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port);
8023 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
8024 if (odp_port != OVSP_NONE) {
8025 ds_put_format(ds, "%"PRIu32":", odp_port);
8027 ds_put_cstr(ds, "none:");
8030 if (strcmp(type, "system")) {
8031 struct netdev *netdev;
8034 ds_put_format(ds, " (%s", type);
8036 error = netdev_open(name, type, &netdev);
8041 error = netdev_get_config(netdev, &config);
8043 const struct smap_node **nodes;
8046 nodes = smap_sort(&config);
8047 for (i = 0; i < smap_count(&config); i++) {
8048 const struct smap_node *node = nodes[i];
8049 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
8050 node->key, node->value);
8054 smap_destroy(&config);
8056 netdev_close(netdev);
8058 ds_put_char(ds, ')');
8060 ds_put_char(ds, '\n');
8066 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
8067 const char *argv[], void *aux OVS_UNUSED)
8069 struct ds ds = DS_EMPTY_INITIALIZER;
8070 const struct ofproto_dpif *ofproto;
8074 for (i = 1; i < argc; i++) {
8075 ofproto = ofproto_dpif_lookup(argv[i]);
8077 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
8078 "for help)", argv[i]);
8079 unixctl_command_reply_error(conn, ds_cstr(&ds));
8082 show_dp_format(ofproto, &ds);
8085 struct shash ofproto_shash;
8086 const struct shash_node **sorted_ofprotos;
8089 shash_init(&ofproto_shash);
8090 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8091 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8092 const struct shash_node *node = sorted_ofprotos[i];
8093 show_dp_format(node->data, &ds);
8096 shash_destroy(&ofproto_shash);
8097 free(sorted_ofprotos);
8100 unixctl_command_reply(conn, ds_cstr(&ds));
8105 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
8106 int argc OVS_UNUSED, const char *argv[],
8107 void *aux OVS_UNUSED)
8109 struct ds ds = DS_EMPTY_INITIALIZER;
8110 const struct ofproto_dpif *ofproto;
8111 struct subfacet *subfacet;
8113 ofproto = ofproto_dpif_lookup(argv[1]);
8115 unixctl_command_reply_error(conn, "no such bridge");
8119 update_stats(ofproto->backer);
8121 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
8122 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
8124 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
8125 subfacet->dp_packet_count, subfacet->dp_byte_count);
8126 if (subfacet->used) {
8127 ds_put_format(&ds, "%.3fs",
8128 (time_msec() - subfacet->used) / 1000.0);
8130 ds_put_format(&ds, "never");
8132 if (subfacet->facet->tcp_flags) {
8133 ds_put_cstr(&ds, ", flags:");
8134 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
8137 ds_put_cstr(&ds, ", actions:");
8138 format_odp_actions(&ds, subfacet->actions, subfacet->actions_len);
8139 ds_put_char(&ds, '\n');
8142 unixctl_command_reply(conn, ds_cstr(&ds));
8147 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
8148 int argc OVS_UNUSED, const char *argv[],
8149 void *aux OVS_UNUSED)
8151 struct ds ds = DS_EMPTY_INITIALIZER;
8152 struct ofproto_dpif *ofproto;
8154 ofproto = ofproto_dpif_lookup(argv[1]);
8156 unixctl_command_reply_error(conn, "no such bridge");
8160 flush(&ofproto->up);
8162 unixctl_command_reply(conn, ds_cstr(&ds));
8167 ofproto_dpif_unixctl_init(void)
8169 static bool registered;
8175 unixctl_command_register(
8177 "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}",
8178 2, 6, ofproto_unixctl_trace, NULL);
8179 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
8180 ofproto_unixctl_fdb_flush, NULL);
8181 unixctl_command_register("fdb/show", "bridge", 1, 1,
8182 ofproto_unixctl_fdb_show, NULL);
8183 unixctl_command_register("ofproto/clog", "", 0, 0,
8184 ofproto_dpif_clog, NULL);
8185 unixctl_command_register("ofproto/unclog", "", 0, 0,
8186 ofproto_dpif_unclog, NULL);
8187 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
8188 ofproto_dpif_self_check, NULL);
8189 unixctl_command_register("dpif/dump-dps", "", 0, 0,
8190 ofproto_unixctl_dpif_dump_dps, NULL);
8191 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
8192 ofproto_unixctl_dpif_show, NULL);
8193 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
8194 ofproto_unixctl_dpif_dump_flows, NULL);
8195 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
8196 ofproto_unixctl_dpif_del_flows, NULL);
8199 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
8201 * This is deprecated. It is only for compatibility with broken device drivers
8202 * in old versions of Linux that do not properly support VLANs when VLAN
8203 * devices are not used. When broken device drivers are no longer in
8204 * widespread use, we will delete these interfaces. */
8207 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
8209 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
8210 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
8212 if (realdev_ofp_port == ofport->realdev_ofp_port
8213 && vid == ofport->vlandev_vid) {
8217 ofproto->backer->need_revalidate = REV_RECONFIGURE;
8219 if (ofport->realdev_ofp_port) {
8222 if (realdev_ofp_port && ofport->bundle) {
8223 /* vlandevs are enslaved to their realdevs, so they are not allowed to
8224 * themselves be part of a bundle. */
8225 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
8228 ofport->realdev_ofp_port = realdev_ofp_port;
8229 ofport->vlandev_vid = vid;
8231 if (realdev_ofp_port) {
8232 vsp_add(ofport, realdev_ofp_port, vid);
8239 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
8241 return hash_2words(realdev_ofp_port, vid);
8244 /* Returns the ODP port number of the Linux VLAN device that corresponds to
8245 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
8246 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
8247 * it would return the port number of eth0.9.
8249 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
8250 * function just returns its 'realdev_odp_port' argument. */
8252 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
8253 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
8255 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
8256 uint16_t realdev_ofp_port;
8257 int vid = vlan_tci_to_vid(vlan_tci);
8258 const struct vlan_splinter *vsp;
8260 realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port);
8261 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
8262 hash_realdev_vid(realdev_ofp_port, vid),
8263 &ofproto->realdev_vid_map) {
8264 if (vsp->realdev_ofp_port == realdev_ofp_port
8265 && vsp->vid == vid) {
8266 return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port);
8270 return realdev_odp_port;
8273 static struct vlan_splinter *
8274 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
8276 struct vlan_splinter *vsp;
8278 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
8279 &ofproto->vlandev_map) {
8280 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
8288 /* Returns the OpenFlow port number of the "real" device underlying the Linux
8289 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
8290 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
8291 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
8292 * eth0 and store 9 in '*vid'.
8294 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
8295 * VLAN device. Unless VLAN splinters are enabled, this is what this function
8298 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
8299 uint16_t vlandev_ofp_port, int *vid)
8301 if (!hmap_is_empty(&ofproto->vlandev_map)) {
8302 const struct vlan_splinter *vsp;
8304 vsp = vlandev_find(ofproto, vlandev_ofp_port);
8309 return vsp->realdev_ofp_port;
8315 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
8316 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
8317 * 'flow->in_port' to the "real" device backing the VLAN device, sets
8318 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
8319 * always the case unless VLAN splinters are enabled), returns false without
8320 * making any changes. */
8322 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
8327 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
8332 /* Cause the flow to be processed as if it came in on the real device with
8333 * the VLAN device's VLAN ID. */
8334 flow->in_port = realdev;
8335 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
8340 vsp_remove(struct ofport_dpif *port)
8342 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8343 struct vlan_splinter *vsp;
8345 vsp = vlandev_find(ofproto, port->up.ofp_port);
8347 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
8348 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
8351 port->realdev_ofp_port = 0;
8353 VLOG_ERR("missing vlan device record");
8358 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
8360 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8362 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
8363 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
8364 == realdev_ofp_port)) {
8365 struct vlan_splinter *vsp;
8367 vsp = xmalloc(sizeof *vsp);
8368 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
8369 hash_int(port->up.ofp_port, 0));
8370 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
8371 hash_realdev_vid(realdev_ofp_port, vid));
8372 vsp->realdev_ofp_port = realdev_ofp_port;
8373 vsp->vlandev_ofp_port = port->up.ofp_port;
8376 port->realdev_ofp_port = realdev_ofp_port;
8378 VLOG_ERR("duplicate vlan device record");
8383 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
8385 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
8386 return ofport ? ofport->odp_port : OVSP_NONE;
8389 static struct ofport_dpif *
8390 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
8392 struct ofport_dpif *port;
8394 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
8395 hash_int(odp_port, 0),
8396 &backer->odp_to_ofport_map) {
8397 if (port->odp_port == odp_port) {
8406 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
8408 struct ofport_dpif *port;
8410 port = odp_port_to_ofport(ofproto->backer, odp_port);
8411 if (port && &ofproto->up == port->up.ofproto) {
8412 return port->up.ofp_port;
8418 const struct ofproto_class ofproto_dpif_class = {
8453 port_is_lacp_current,
8454 NULL, /* rule_choose_table */
8461 rule_modify_actions,
8472 get_stp_port_status,
8479 is_mirror_output_bundle,
8480 forward_bpdu_changed,
8481 set_mac_table_config,