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. However, 'one_subfacet' may not
491 * always be valid, since it could have been removed after newer
492 * subfacets were pushed onto the 'subfacets' list.) */
493 struct subfacet one_subfacet;
496 static struct facet *facet_create(struct rule_dpif *,
497 const struct flow *, uint32_t hash);
498 static void facet_remove(struct facet *);
499 static void facet_free(struct facet *);
501 static struct facet *facet_find(struct ofproto_dpif *,
502 const struct flow *, uint32_t hash);
503 static struct facet *facet_lookup_valid(struct ofproto_dpif *,
504 const struct flow *, uint32_t hash);
505 static void facet_revalidate(struct facet *);
506 static bool facet_check_consistency(struct facet *);
508 static void facet_flush_stats(struct facet *);
510 static void facet_update_time(struct facet *, long long int used);
511 static void facet_reset_counters(struct facet *);
512 static void facet_push_stats(struct facet *);
513 static void facet_learn(struct facet *);
514 static void facet_account(struct facet *);
516 static struct subfacet *facet_get_subfacet(struct facet *);
518 static bool facet_is_controller_flow(struct facet *);
521 struct hmap_node odp_port_node; /* In dpif_backer's "odp_to_ofport_map". */
525 struct ofbundle *bundle; /* Bundle that contains this port, if any. */
526 struct list bundle_node; /* In struct ofbundle's "ports" list. */
527 struct cfm *cfm; /* Connectivity Fault Management, if any. */
528 tag_type tag; /* Tag associated with this port. */
529 bool may_enable; /* May be enabled in bonds. */
530 long long int carrier_seq; /* Carrier status changes. */
531 struct tnl_port *tnl_port; /* Tunnel handle, or null. */
534 struct stp_port *stp_port; /* Spanning Tree Protocol, if any. */
535 enum stp_state stp_state; /* Always STP_DISABLED if STP not in use. */
536 long long int stp_state_entered;
538 struct hmap priorities; /* Map of attached 'priority_to_dscp's. */
540 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
542 * This is deprecated. It is only for compatibility with broken device
543 * drivers in old versions of Linux that do not properly support VLANs when
544 * VLAN devices are not used. When broken device drivers are no longer in
545 * widespread use, we will delete these interfaces. */
546 uint16_t realdev_ofp_port;
550 /* Node in 'ofport_dpif''s 'priorities' map. Used to maintain a map from
551 * 'priority' (the datapath's term for QoS queue) to the dscp bits which all
552 * traffic egressing the 'ofport' with that priority should be marked with. */
553 struct priority_to_dscp {
554 struct hmap_node hmap_node; /* Node in 'ofport_dpif''s 'priorities' map. */
555 uint32_t priority; /* Priority of this queue (see struct flow). */
557 uint8_t dscp; /* DSCP bits to mark outgoing traffic with. */
560 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
562 * This is deprecated. It is only for compatibility with broken device drivers
563 * in old versions of Linux that do not properly support VLANs when VLAN
564 * devices are not used. When broken device drivers are no longer in
565 * widespread use, we will delete these interfaces. */
566 struct vlan_splinter {
567 struct hmap_node realdev_vid_node;
568 struct hmap_node vlandev_node;
569 uint16_t realdev_ofp_port;
570 uint16_t vlandev_ofp_port;
574 static uint32_t vsp_realdev_to_vlandev(const struct ofproto_dpif *,
575 uint32_t realdev, ovs_be16 vlan_tci);
576 static bool vsp_adjust_flow(const struct ofproto_dpif *, struct flow *);
577 static void vsp_remove(struct ofport_dpif *);
578 static void vsp_add(struct ofport_dpif *, uint16_t realdev_ofp_port, int vid);
580 static uint32_t ofp_port_to_odp_port(const struct ofproto_dpif *,
582 static uint16_t odp_port_to_ofp_port(const struct ofproto_dpif *,
585 static struct ofport_dpif *
586 ofport_dpif_cast(const struct ofport *ofport)
588 ovs_assert(ofport->ofproto->ofproto_class == &ofproto_dpif_class);
589 return ofport ? CONTAINER_OF(ofport, struct ofport_dpif, up) : NULL;
592 static void port_run(struct ofport_dpif *);
593 static void port_run_fast(struct ofport_dpif *);
594 static void port_wait(struct ofport_dpif *);
595 static int set_cfm(struct ofport *, const struct cfm_settings *);
596 static void ofport_clear_priorities(struct ofport_dpif *);
598 struct dpif_completion {
599 struct list list_node;
600 struct ofoperation *op;
603 /* Extra information about a classifier table.
604 * Currently used just for optimized flow revalidation. */
606 /* If either of these is nonnull, then this table has a form that allows
607 * flows to be tagged to avoid revalidating most flows for the most common
608 * kinds of flow table changes. */
609 struct cls_table *catchall_table; /* Table that wildcards all fields. */
610 struct cls_table *other_table; /* Table with any other wildcard set. */
611 uint32_t basis; /* Keeps each table's tags separate. */
614 /* Reasons that we might need to revalidate every facet, and corresponding
617 * A value of 0 means that there is no need to revalidate.
619 * It would be nice to have some cleaner way to integrate with coverage
620 * counters, but with only a few reasons I guess this is good enough for
622 enum revalidate_reason {
623 REV_RECONFIGURE = 1, /* Switch configuration changed. */
624 REV_STP, /* Spanning tree protocol port status change. */
625 REV_PORT_TOGGLED, /* Port enabled or disabled by CFM, LACP, ...*/
626 REV_FLOW_TABLE, /* Flow table changed. */
627 REV_INCONSISTENCY /* Facet self-check failed. */
629 COVERAGE_DEFINE(rev_reconfigure);
630 COVERAGE_DEFINE(rev_stp);
631 COVERAGE_DEFINE(rev_port_toggled);
632 COVERAGE_DEFINE(rev_flow_table);
633 COVERAGE_DEFINE(rev_inconsistency);
635 /* Drop keys are odp flow keys which have drop flows installed in the kernel.
636 * These are datapath flows which have no associated ofproto, if they did we
637 * would use facets. */
639 struct hmap_node hmap_node;
644 /* All datapaths of a given type share a single dpif backer instance. */
649 struct timer next_expiration;
650 struct hmap odp_to_ofport_map; /* ODP port to ofport mapping. */
652 struct simap tnl_backers; /* Set of dpif ports backing tunnels. */
654 /* Facet revalidation flags applying to facets which use this backer. */
655 enum revalidate_reason need_revalidate; /* Revalidate every facet. */
656 struct tag_set revalidate_set; /* Revalidate only matching facets. */
658 struct hmap drop_keys; /* Set of dropped odp keys. */
661 /* All existing ofproto_backer instances, indexed by ofproto->up.type. */
662 static struct shash all_dpif_backers = SHASH_INITIALIZER(&all_dpif_backers);
664 static void drop_key_clear(struct dpif_backer *);
665 static struct ofport_dpif *
666 odp_port_to_ofport(const struct dpif_backer *, uint32_t odp_port);
668 static void dpif_stats_update_hit_count(struct ofproto_dpif *ofproto,
671 struct ofproto_dpif {
672 struct hmap_node all_ofproto_dpifs_node; /* In 'all_ofproto_dpifs'. */
674 struct dpif_backer *backer;
676 /* Special OpenFlow rules. */
677 struct rule_dpif *miss_rule; /* Sends flow table misses to controller. */
678 struct rule_dpif *no_packet_in_rule; /* Drops flow table misses. */
684 struct netflow *netflow;
685 struct dpif_sflow *sflow;
686 struct hmap bundles; /* Contains "struct ofbundle"s. */
687 struct mac_learning *ml;
688 struct ofmirror *mirrors[MAX_MIRRORS];
690 bool has_bonded_bundles;
694 struct hmap subfacets;
695 struct governor *governor;
698 struct table_dpif tables[N_TABLES];
700 /* Support for debugging async flow mods. */
701 struct list completions;
703 bool has_bundle_action; /* True when the first bundle action appears. */
704 struct netdev_stats stats; /* To account packets generated and consumed in
709 long long int stp_last_tick;
711 /* VLAN splinters. */
712 struct hmap realdev_vid_map; /* (realdev,vid) -> vlandev. */
713 struct hmap vlandev_map; /* vlandev -> (realdev,vid). */
716 struct sset ports; /* Set of standard port names. */
717 struct sset ghost_ports; /* Ports with no datapath port. */
718 struct sset port_poll_set; /* Queued names for port_poll() reply. */
719 int port_poll_errno; /* Last errno for port_poll() reply. */
721 /* Per ofproto's dpif stats. */
726 /* Defer flow mod completion until "ovs-appctl ofproto/unclog"? (Useful only
727 * for debugging the asynchronous flow_mod implementation.) */
730 /* All existing ofproto_dpif instances, indexed by ->up.name. */
731 static struct hmap all_ofproto_dpifs = HMAP_INITIALIZER(&all_ofproto_dpifs);
733 static void ofproto_dpif_unixctl_init(void);
735 static struct ofproto_dpif *
736 ofproto_dpif_cast(const struct ofproto *ofproto)
738 ovs_assert(ofproto->ofproto_class == &ofproto_dpif_class);
739 return CONTAINER_OF(ofproto, struct ofproto_dpif, up);
742 static struct ofport_dpif *get_ofp_port(const struct ofproto_dpif *,
744 static struct ofport_dpif *get_odp_port(const struct ofproto_dpif *,
746 static void ofproto_trace(struct ofproto_dpif *, const struct flow *,
747 const struct ofpbuf *,
748 const struct initial_vals *, struct ds *);
750 /* Packet processing. */
751 static void update_learning_table(struct ofproto_dpif *,
752 const struct flow *, int vlan,
755 #define FLOW_MISS_MAX_BATCH 50
756 static int handle_upcalls(struct dpif_backer *, unsigned int max_batch);
758 /* Flow expiration. */
759 static int expire(struct dpif_backer *);
762 static void send_netflow_active_timeouts(struct ofproto_dpif *);
765 static int send_packet(const struct ofport_dpif *, struct ofpbuf *packet);
766 static size_t compose_sflow_action(const struct ofproto_dpif *,
767 struct ofpbuf *odp_actions,
768 const struct flow *, uint32_t odp_port);
769 static void add_mirror_actions(struct action_xlate_ctx *ctx,
770 const struct flow *flow);
771 /* Global variables. */
772 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
774 /* Initial mappings of port to bridge mappings. */
775 static struct shash init_ofp_ports = SHASH_INITIALIZER(&init_ofp_ports);
777 /* Factory functions. */
780 init(const struct shash *iface_hints)
782 struct shash_node *node;
784 /* Make a local copy, since we don't own 'iface_hints' elements. */
785 SHASH_FOR_EACH(node, iface_hints) {
786 const struct iface_hint *orig_hint = node->data;
787 struct iface_hint *new_hint = xmalloc(sizeof *new_hint);
789 new_hint->br_name = xstrdup(orig_hint->br_name);
790 new_hint->br_type = xstrdup(orig_hint->br_type);
791 new_hint->ofp_port = orig_hint->ofp_port;
793 shash_add(&init_ofp_ports, node->name, new_hint);
798 enumerate_types(struct sset *types)
800 dp_enumerate_types(types);
804 enumerate_names(const char *type, struct sset *names)
806 struct ofproto_dpif *ofproto;
809 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
810 if (strcmp(type, ofproto->up.type)) {
813 sset_add(names, ofproto->up.name);
820 del(const char *type, const char *name)
825 error = dpif_open(name, type, &dpif);
827 error = dpif_delete(dpif);
834 port_open_type(const char *datapath_type, const char *port_type)
836 return dpif_port_open_type(datapath_type, port_type);
839 /* Type functions. */
841 static struct ofproto_dpif *
842 lookup_ofproto_dpif_by_port_name(const char *name)
844 struct ofproto_dpif *ofproto;
846 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
847 if (sset_contains(&ofproto->ports, name)) {
856 type_run(const char *type)
858 struct dpif_backer *backer;
862 backer = shash_find_data(&all_dpif_backers, type);
864 /* This is not necessarily a problem, since backers are only
865 * created on demand. */
869 dpif_run(backer->dpif);
871 if (backer->need_revalidate
872 || !tag_set_is_empty(&backer->revalidate_set)) {
873 struct tag_set revalidate_set = backer->revalidate_set;
874 bool need_revalidate = backer->need_revalidate;
875 struct ofproto_dpif *ofproto;
876 struct simap_node *node;
877 struct simap tmp_backers;
879 /* Handle tunnel garbage collection. */
880 simap_init(&tmp_backers);
881 simap_swap(&backer->tnl_backers, &tmp_backers);
883 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
884 struct ofport_dpif *iter;
886 if (backer != ofproto->backer) {
890 HMAP_FOR_EACH (iter, up.hmap_node, &ofproto->up.ports) {
893 if (!iter->tnl_port) {
897 dp_port = netdev_vport_get_dpif_port(iter->up.netdev);
898 node = simap_find(&tmp_backers, dp_port);
900 simap_put(&backer->tnl_backers, dp_port, node->data);
901 simap_delete(&tmp_backers, node);
902 node = simap_find(&backer->tnl_backers, dp_port);
904 node = simap_find(&backer->tnl_backers, dp_port);
906 uint32_t odp_port = UINT32_MAX;
908 if (!dpif_port_add(backer->dpif, iter->up.netdev,
910 simap_put(&backer->tnl_backers, dp_port, odp_port);
911 node = simap_find(&backer->tnl_backers, dp_port);
916 iter->odp_port = node ? node->data : OVSP_NONE;
917 if (tnl_port_reconfigure(&iter->up, iter->odp_port,
919 backer->need_revalidate = REV_RECONFIGURE;
924 SIMAP_FOR_EACH (node, &tmp_backers) {
925 dpif_port_del(backer->dpif, node->data);
927 simap_destroy(&tmp_backers);
929 switch (backer->need_revalidate) {
930 case REV_RECONFIGURE: COVERAGE_INC(rev_reconfigure); break;
931 case REV_STP: COVERAGE_INC(rev_stp); break;
932 case REV_PORT_TOGGLED: COVERAGE_INC(rev_port_toggled); break;
933 case REV_FLOW_TABLE: COVERAGE_INC(rev_flow_table); break;
934 case REV_INCONSISTENCY: COVERAGE_INC(rev_inconsistency); break;
937 if (backer->need_revalidate) {
938 /* Clear the drop_keys in case we should now be accepting some
939 * formerly dropped flows. */
940 drop_key_clear(backer);
943 /* Clear the revalidation flags. */
944 tag_set_init(&backer->revalidate_set);
945 backer->need_revalidate = 0;
947 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
948 struct facet *facet, *next;
950 if (ofproto->backer != backer) {
954 HMAP_FOR_EACH_SAFE (facet, next, hmap_node, &ofproto->facets) {
956 || tag_set_intersects(&revalidate_set, facet->tags)) {
957 facet_revalidate(facet);
963 if (timer_expired(&backer->next_expiration)) {
964 int delay = expire(backer);
965 timer_set_duration(&backer->next_expiration, delay);
968 /* Check for port changes in the dpif. */
969 while ((error = dpif_port_poll(backer->dpif, &devname)) == 0) {
970 struct ofproto_dpif *ofproto;
971 struct dpif_port port;
973 /* Don't report on the datapath's device. */
974 if (!strcmp(devname, dpif_base_name(backer->dpif))) {
978 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
979 &all_ofproto_dpifs) {
980 if (simap_contains(&ofproto->backer->tnl_backers, devname)) {
985 ofproto = lookup_ofproto_dpif_by_port_name(devname);
986 if (dpif_port_query_by_name(backer->dpif, devname, &port)) {
987 /* The port was removed. If we know the datapath,
988 * report it through poll_set(). If we don't, it may be
989 * notifying us of a removal we initiated, so ignore it.
990 * If there's a pending ENOBUFS, let it stand, since
991 * everything will be reevaluated. */
992 if (ofproto && ofproto->port_poll_errno != ENOBUFS) {
993 sset_add(&ofproto->port_poll_set, devname);
994 ofproto->port_poll_errno = 0;
996 } else if (!ofproto) {
997 /* The port was added, but we don't know with which
998 * ofproto we should associate it. Delete it. */
999 dpif_port_del(backer->dpif, port.port_no);
1001 dpif_port_destroy(&port);
1007 if (error != EAGAIN) {
1008 struct ofproto_dpif *ofproto;
1010 /* There was some sort of error, so propagate it to all
1011 * ofprotos that use this backer. */
1012 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node,
1013 &all_ofproto_dpifs) {
1014 if (ofproto->backer == backer) {
1015 sset_clear(&ofproto->port_poll_set);
1016 ofproto->port_poll_errno = error;
1025 type_run_fast(const char *type)
1027 struct dpif_backer *backer;
1030 backer = shash_find_data(&all_dpif_backers, type);
1032 /* This is not necessarily a problem, since backers are only
1033 * created on demand. */
1037 /* Handle one or more batches of upcalls, until there's nothing left to do
1038 * or until we do a fixed total amount of work.
1040 * We do work in batches because it can be much cheaper to set up a number
1041 * of flows and fire off their patches all at once. We do multiple batches
1042 * because in some cases handling a packet can cause another packet to be
1043 * queued almost immediately as part of the return flow. Both
1044 * optimizations can make major improvements on some benchmarks and
1045 * presumably for real traffic as well. */
1047 while (work < FLOW_MISS_MAX_BATCH) {
1048 int retval = handle_upcalls(backer, FLOW_MISS_MAX_BATCH - work);
1059 type_wait(const char *type)
1061 struct dpif_backer *backer;
1063 backer = shash_find_data(&all_dpif_backers, type);
1065 /* This is not necessarily a problem, since backers are only
1066 * created on demand. */
1070 timer_wait(&backer->next_expiration);
1073 /* Basic life-cycle. */
1075 static int add_internal_flows(struct ofproto_dpif *);
1077 static struct ofproto *
1080 struct ofproto_dpif *ofproto = xmalloc(sizeof *ofproto);
1081 return &ofproto->up;
1085 dealloc(struct ofproto *ofproto_)
1087 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1092 close_dpif_backer(struct dpif_backer *backer)
1094 struct shash_node *node;
1096 ovs_assert(backer->refcount > 0);
1098 if (--backer->refcount) {
1102 drop_key_clear(backer);
1103 hmap_destroy(&backer->drop_keys);
1105 simap_destroy(&backer->tnl_backers);
1106 hmap_destroy(&backer->odp_to_ofport_map);
1107 node = shash_find(&all_dpif_backers, backer->type);
1109 shash_delete(&all_dpif_backers, node);
1110 dpif_close(backer->dpif);
1115 /* Datapath port slated for removal from datapath. */
1116 struct odp_garbage {
1117 struct list list_node;
1122 open_dpif_backer(const char *type, struct dpif_backer **backerp)
1124 struct dpif_backer *backer;
1125 struct dpif_port_dump port_dump;
1126 struct dpif_port port;
1127 struct shash_node *node;
1128 struct list garbage_list;
1129 struct odp_garbage *garbage, *next;
1135 backer = shash_find_data(&all_dpif_backers, type);
1142 backer_name = xasprintf("ovs-%s", type);
1144 /* Remove any existing datapaths, since we assume we're the only
1145 * userspace controlling the datapath. */
1147 dp_enumerate_names(type, &names);
1148 SSET_FOR_EACH(name, &names) {
1149 struct dpif *old_dpif;
1151 /* Don't remove our backer if it exists. */
1152 if (!strcmp(name, backer_name)) {
1156 if (dpif_open(name, type, &old_dpif)) {
1157 VLOG_WARN("couldn't open old datapath %s to remove it", name);
1159 dpif_delete(old_dpif);
1160 dpif_close(old_dpif);
1163 sset_destroy(&names);
1165 backer = xmalloc(sizeof *backer);
1167 error = dpif_create_and_open(backer_name, type, &backer->dpif);
1170 VLOG_ERR("failed to open datapath of type %s: %s", type,
1176 backer->type = xstrdup(type);
1177 backer->refcount = 1;
1178 hmap_init(&backer->odp_to_ofport_map);
1179 hmap_init(&backer->drop_keys);
1180 timer_set_duration(&backer->next_expiration, 1000);
1181 backer->need_revalidate = 0;
1182 simap_init(&backer->tnl_backers);
1183 tag_set_init(&backer->revalidate_set);
1186 dpif_flow_flush(backer->dpif);
1188 /* Loop through the ports already on the datapath and remove any
1189 * that we don't need anymore. */
1190 list_init(&garbage_list);
1191 dpif_port_dump_start(&port_dump, backer->dpif);
1192 while (dpif_port_dump_next(&port_dump, &port)) {
1193 node = shash_find(&init_ofp_ports, port.name);
1194 if (!node && strcmp(port.name, dpif_base_name(backer->dpif))) {
1195 garbage = xmalloc(sizeof *garbage);
1196 garbage->odp_port = port.port_no;
1197 list_push_front(&garbage_list, &garbage->list_node);
1200 dpif_port_dump_done(&port_dump);
1202 LIST_FOR_EACH_SAFE (garbage, next, list_node, &garbage_list) {
1203 dpif_port_del(backer->dpif, garbage->odp_port);
1204 list_remove(&garbage->list_node);
1208 shash_add(&all_dpif_backers, type, backer);
1210 error = dpif_recv_set(backer->dpif, true);
1212 VLOG_ERR("failed to listen on datapath of type %s: %s",
1213 type, strerror(error));
1214 close_dpif_backer(backer);
1222 construct(struct ofproto *ofproto_)
1224 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1225 struct shash_node *node, *next;
1230 error = open_dpif_backer(ofproto->up.type, &ofproto->backer);
1235 max_ports = dpif_get_max_ports(ofproto->backer->dpif);
1236 ofproto_init_max_ports(ofproto_, MIN(max_ports, OFPP_MAX));
1238 ofproto->n_matches = 0;
1240 ofproto->netflow = NULL;
1241 ofproto->sflow = NULL;
1242 ofproto->stp = NULL;
1243 hmap_init(&ofproto->bundles);
1244 ofproto->ml = mac_learning_create(MAC_ENTRY_DEFAULT_IDLE_TIME);
1245 for (i = 0; i < MAX_MIRRORS; i++) {
1246 ofproto->mirrors[i] = NULL;
1248 ofproto->has_bonded_bundles = false;
1250 hmap_init(&ofproto->facets);
1251 hmap_init(&ofproto->subfacets);
1252 ofproto->governor = NULL;
1254 for (i = 0; i < N_TABLES; i++) {
1255 struct table_dpif *table = &ofproto->tables[i];
1257 table->catchall_table = NULL;
1258 table->other_table = NULL;
1259 table->basis = random_uint32();
1262 list_init(&ofproto->completions);
1264 ofproto_dpif_unixctl_init();
1266 ofproto->has_mirrors = false;
1267 ofproto->has_bundle_action = false;
1269 hmap_init(&ofproto->vlandev_map);
1270 hmap_init(&ofproto->realdev_vid_map);
1272 sset_init(&ofproto->ports);
1273 sset_init(&ofproto->ghost_ports);
1274 sset_init(&ofproto->port_poll_set);
1275 ofproto->port_poll_errno = 0;
1277 SHASH_FOR_EACH_SAFE (node, next, &init_ofp_ports) {
1278 struct iface_hint *iface_hint = node->data;
1280 if (!strcmp(iface_hint->br_name, ofproto->up.name)) {
1281 /* Check if the datapath already has this port. */
1282 if (dpif_port_exists(ofproto->backer->dpif, node->name)) {
1283 sset_add(&ofproto->ports, node->name);
1286 free(iface_hint->br_name);
1287 free(iface_hint->br_type);
1289 shash_delete(&init_ofp_ports, node);
1293 hmap_insert(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node,
1294 hash_string(ofproto->up.name, 0));
1295 memset(&ofproto->stats, 0, sizeof ofproto->stats);
1297 ofproto_init_tables(ofproto_, N_TABLES);
1298 error = add_internal_flows(ofproto);
1299 ofproto->up.tables[TBL_INTERNAL].flags = OFTABLE_HIDDEN | OFTABLE_READONLY;
1302 ofproto->n_missed = 0;
1308 add_internal_flow(struct ofproto_dpif *ofproto, int id,
1309 const struct ofpbuf *ofpacts, struct rule_dpif **rulep)
1311 struct ofputil_flow_mod fm;
1314 match_init_catchall(&fm.match);
1316 match_set_reg(&fm.match, 0, id);
1317 fm.new_cookie = htonll(0);
1318 fm.cookie = htonll(0);
1319 fm.cookie_mask = htonll(0);
1320 fm.table_id = TBL_INTERNAL;
1321 fm.command = OFPFC_ADD;
1322 fm.idle_timeout = 0;
1323 fm.hard_timeout = 0;
1327 fm.ofpacts = ofpacts->data;
1328 fm.ofpacts_len = ofpacts->size;
1330 error = ofproto_flow_mod(&ofproto->up, &fm);
1332 VLOG_ERR_RL(&rl, "failed to add internal flow %d (%s)",
1333 id, ofperr_to_string(error));
1337 *rulep = rule_dpif_lookup__(ofproto, &fm.match.flow, TBL_INTERNAL);
1338 ovs_assert(*rulep != NULL);
1344 add_internal_flows(struct ofproto_dpif *ofproto)
1346 struct ofpact_controller *controller;
1347 uint64_t ofpacts_stub[128 / 8];
1348 struct ofpbuf ofpacts;
1352 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
1355 controller = ofpact_put_CONTROLLER(&ofpacts);
1356 controller->max_len = UINT16_MAX;
1357 controller->controller_id = 0;
1358 controller->reason = OFPR_NO_MATCH;
1359 ofpact_pad(&ofpacts);
1361 error = add_internal_flow(ofproto, id++, &ofpacts, &ofproto->miss_rule);
1366 ofpbuf_clear(&ofpacts);
1367 error = add_internal_flow(ofproto, id++, &ofpacts,
1368 &ofproto->no_packet_in_rule);
1373 complete_operations(struct ofproto_dpif *ofproto)
1375 struct dpif_completion *c, *next;
1377 LIST_FOR_EACH_SAFE (c, next, list_node, &ofproto->completions) {
1378 ofoperation_complete(c->op, 0);
1379 list_remove(&c->list_node);
1385 destruct(struct ofproto *ofproto_)
1387 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1388 struct rule_dpif *rule, *next_rule;
1389 struct oftable *table;
1392 hmap_remove(&all_ofproto_dpifs, &ofproto->all_ofproto_dpifs_node);
1393 complete_operations(ofproto);
1395 OFPROTO_FOR_EACH_TABLE (table, &ofproto->up) {
1396 struct cls_cursor cursor;
1398 cls_cursor_init(&cursor, &table->cls, NULL);
1399 CLS_CURSOR_FOR_EACH_SAFE (rule, next_rule, up.cr, &cursor) {
1400 ofproto_rule_destroy(&rule->up);
1404 for (i = 0; i < MAX_MIRRORS; i++) {
1405 mirror_destroy(ofproto->mirrors[i]);
1408 netflow_destroy(ofproto->netflow);
1409 dpif_sflow_destroy(ofproto->sflow);
1410 hmap_destroy(&ofproto->bundles);
1411 mac_learning_destroy(ofproto->ml);
1413 hmap_destroy(&ofproto->facets);
1414 hmap_destroy(&ofproto->subfacets);
1415 governor_destroy(ofproto->governor);
1417 hmap_destroy(&ofproto->vlandev_map);
1418 hmap_destroy(&ofproto->realdev_vid_map);
1420 sset_destroy(&ofproto->ports);
1421 sset_destroy(&ofproto->ghost_ports);
1422 sset_destroy(&ofproto->port_poll_set);
1424 close_dpif_backer(ofproto->backer);
1428 run_fast(struct ofproto *ofproto_)
1430 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1431 struct ofport_dpif *ofport;
1433 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1434 port_run_fast(ofport);
1441 run(struct ofproto *ofproto_)
1443 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1444 struct ofport_dpif *ofport;
1445 struct ofbundle *bundle;
1449 complete_operations(ofproto);
1452 error = run_fast(ofproto_);
1457 if (ofproto->netflow) {
1458 if (netflow_run(ofproto->netflow)) {
1459 send_netflow_active_timeouts(ofproto);
1462 if (ofproto->sflow) {
1463 dpif_sflow_run(ofproto->sflow);
1466 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1469 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1474 mac_learning_run(ofproto->ml, &ofproto->backer->revalidate_set);
1476 /* Check the consistency of a random facet, to aid debugging. */
1477 if (!hmap_is_empty(&ofproto->facets)
1478 && !ofproto->backer->need_revalidate) {
1479 struct facet *facet;
1481 facet = CONTAINER_OF(hmap_random_node(&ofproto->facets),
1482 struct facet, hmap_node);
1483 if (!tag_set_intersects(&ofproto->backer->revalidate_set,
1485 if (!facet_check_consistency(facet)) {
1486 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
1491 if (ofproto->governor) {
1494 governor_run(ofproto->governor);
1496 /* If the governor has shrunk to its minimum size and the number of
1497 * subfacets has dwindled, then drop the governor entirely.
1499 * For hysteresis, the number of subfacets to drop the governor is
1500 * smaller than the number needed to trigger its creation. */
1501 n_subfacets = hmap_count(&ofproto->subfacets);
1502 if (n_subfacets * 4 < ofproto->up.flow_eviction_threshold
1503 && governor_is_idle(ofproto->governor)) {
1504 governor_destroy(ofproto->governor);
1505 ofproto->governor = NULL;
1513 wait(struct ofproto *ofproto_)
1515 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1516 struct ofport_dpif *ofport;
1517 struct ofbundle *bundle;
1519 if (!clogged && !list_is_empty(&ofproto->completions)) {
1520 poll_immediate_wake();
1523 dpif_wait(ofproto->backer->dpif);
1524 dpif_recv_wait(ofproto->backer->dpif);
1525 if (ofproto->sflow) {
1526 dpif_sflow_wait(ofproto->sflow);
1528 if (!tag_set_is_empty(&ofproto->backer->revalidate_set)) {
1529 poll_immediate_wake();
1531 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1534 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
1535 bundle_wait(bundle);
1537 if (ofproto->netflow) {
1538 netflow_wait(ofproto->netflow);
1540 mac_learning_wait(ofproto->ml);
1542 if (ofproto->backer->need_revalidate) {
1543 /* Shouldn't happen, but if it does just go around again. */
1544 VLOG_DBG_RL(&rl, "need revalidate in ofproto_wait_cb()");
1545 poll_immediate_wake();
1547 if (ofproto->governor) {
1548 governor_wait(ofproto->governor);
1553 get_memory_usage(const struct ofproto *ofproto_, struct simap *usage)
1555 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1557 simap_increase(usage, "facets", hmap_count(&ofproto->facets));
1558 simap_increase(usage, "subfacets", hmap_count(&ofproto->subfacets));
1562 flush(struct ofproto *ofproto_)
1564 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1565 struct subfacet *subfacet, *next_subfacet;
1566 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
1570 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
1571 &ofproto->subfacets) {
1572 if (subfacet->path != SF_NOT_INSTALLED) {
1573 batch[n_batch++] = subfacet;
1574 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
1575 subfacet_destroy_batch(ofproto, batch, n_batch);
1579 subfacet_destroy(subfacet);
1584 subfacet_destroy_batch(ofproto, batch, n_batch);
1589 get_features(struct ofproto *ofproto_ OVS_UNUSED,
1590 bool *arp_match_ip, enum ofputil_action_bitmap *actions)
1592 *arp_match_ip = true;
1593 *actions = (OFPUTIL_A_OUTPUT |
1594 OFPUTIL_A_SET_VLAN_VID |
1595 OFPUTIL_A_SET_VLAN_PCP |
1596 OFPUTIL_A_STRIP_VLAN |
1597 OFPUTIL_A_SET_DL_SRC |
1598 OFPUTIL_A_SET_DL_DST |
1599 OFPUTIL_A_SET_NW_SRC |
1600 OFPUTIL_A_SET_NW_DST |
1601 OFPUTIL_A_SET_NW_TOS |
1602 OFPUTIL_A_SET_TP_SRC |
1603 OFPUTIL_A_SET_TP_DST |
1608 get_tables(struct ofproto *ofproto_, struct ofp12_table_stats *ots)
1610 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1611 struct dpif_dp_stats s;
1613 strcpy(ots->name, "classifier");
1615 dpif_get_dp_stats(ofproto->backer->dpif, &s);
1617 ots->lookup_count = htonll(s.n_hit + s.n_missed);
1618 ots->matched_count = htonll(s.n_hit + ofproto->n_matches);
1621 static struct ofport *
1624 struct ofport_dpif *port = xmalloc(sizeof *port);
1629 port_dealloc(struct ofport *port_)
1631 struct ofport_dpif *port = ofport_dpif_cast(port_);
1636 port_construct(struct ofport *port_)
1638 struct ofport_dpif *port = ofport_dpif_cast(port_);
1639 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1640 const struct netdev *netdev = port->up.netdev;
1641 struct dpif_port dpif_port;
1644 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1645 port->bundle = NULL;
1647 port->tag = tag_create_random();
1648 port->may_enable = true;
1649 port->stp_port = NULL;
1650 port->stp_state = STP_DISABLED;
1651 port->tnl_port = NULL;
1652 hmap_init(&port->priorities);
1653 port->realdev_ofp_port = 0;
1654 port->vlandev_vid = 0;
1655 port->carrier_seq = netdev_get_carrier_resets(netdev);
1657 if (netdev_vport_is_patch(netdev)) {
1658 /* XXX By bailing out here, we don't do required sFlow work. */
1659 port->odp_port = OVSP_NONE;
1663 error = dpif_port_query_by_name(ofproto->backer->dpif,
1664 netdev_vport_get_dpif_port(netdev),
1670 port->odp_port = dpif_port.port_no;
1672 if (netdev_get_tunnel_config(netdev)) {
1673 port->tnl_port = tnl_port_add(&port->up, port->odp_port);
1675 /* Sanity-check that a mapping doesn't already exist. This
1676 * shouldn't happen for non-tunnel ports. */
1677 if (odp_port_to_ofp_port(ofproto, port->odp_port) != OFPP_NONE) {
1678 VLOG_ERR("port %s already has an OpenFlow port number",
1680 dpif_port_destroy(&dpif_port);
1684 hmap_insert(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node,
1685 hash_int(port->odp_port, 0));
1687 dpif_port_destroy(&dpif_port);
1689 if (ofproto->sflow) {
1690 dpif_sflow_add_port(ofproto->sflow, port_, port->odp_port);
1697 port_destruct(struct ofport *port_)
1699 struct ofport_dpif *port = ofport_dpif_cast(port_);
1700 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1701 const char *dp_port_name = netdev_vport_get_dpif_port(port->up.netdev);
1702 const char *devname = netdev_get_name(port->up.netdev);
1704 if (dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
1705 /* The underlying device is still there, so delete it. This
1706 * happens when the ofproto is being destroyed, since the caller
1707 * assumes that removal of attached ports will happen as part of
1709 if (!port->tnl_port) {
1710 dpif_port_del(ofproto->backer->dpif, port->odp_port);
1712 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1715 if (port->odp_port != OVSP_NONE && !port->tnl_port) {
1716 hmap_remove(&ofproto->backer->odp_to_ofport_map, &port->odp_port_node);
1719 tnl_port_del(port->tnl_port);
1720 sset_find_and_delete(&ofproto->ports, devname);
1721 sset_find_and_delete(&ofproto->ghost_ports, devname);
1722 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1723 bundle_remove(port_);
1724 set_cfm(port_, NULL);
1725 if (ofproto->sflow) {
1726 dpif_sflow_del_port(ofproto->sflow, port->odp_port);
1729 ofport_clear_priorities(port);
1730 hmap_destroy(&port->priorities);
1734 port_modified(struct ofport *port_)
1736 struct ofport_dpif *port = ofport_dpif_cast(port_);
1738 if (port->bundle && port->bundle->bond) {
1739 bond_slave_set_netdev(port->bundle->bond, port, port->up.netdev);
1744 port_reconfigured(struct ofport *port_, enum ofputil_port_config old_config)
1746 struct ofport_dpif *port = ofport_dpif_cast(port_);
1747 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
1748 enum ofputil_port_config changed = old_config ^ port->up.pp.config;
1750 if (changed & (OFPUTIL_PC_NO_RECV | OFPUTIL_PC_NO_RECV_STP |
1751 OFPUTIL_PC_NO_FWD | OFPUTIL_PC_NO_FLOOD |
1752 OFPUTIL_PC_NO_PACKET_IN)) {
1753 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1755 if (changed & OFPUTIL_PC_NO_FLOOD && port->bundle) {
1756 bundle_update(port->bundle);
1762 set_sflow(struct ofproto *ofproto_,
1763 const struct ofproto_sflow_options *sflow_options)
1765 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1766 struct dpif_sflow *ds = ofproto->sflow;
1768 if (sflow_options) {
1770 struct ofport_dpif *ofport;
1772 ds = ofproto->sflow = dpif_sflow_create();
1773 HMAP_FOR_EACH (ofport, up.hmap_node, &ofproto->up.ports) {
1774 dpif_sflow_add_port(ds, &ofport->up, ofport->odp_port);
1776 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1778 dpif_sflow_set_options(ds, sflow_options);
1781 dpif_sflow_destroy(ds);
1782 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1783 ofproto->sflow = NULL;
1790 set_cfm(struct ofport *ofport_, const struct cfm_settings *s)
1792 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1799 struct ofproto_dpif *ofproto;
1801 ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1802 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1803 ofport->cfm = cfm_create(netdev_get_name(ofport->up.netdev));
1806 if (cfm_configure(ofport->cfm, s)) {
1812 cfm_destroy(ofport->cfm);
1818 get_cfm_status(const struct ofport *ofport_,
1819 struct ofproto_cfm_status *status)
1821 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1824 status->faults = cfm_get_fault(ofport->cfm);
1825 status->remote_opstate = cfm_get_opup(ofport->cfm);
1826 status->health = cfm_get_health(ofport->cfm);
1827 cfm_get_remote_mpids(ofport->cfm, &status->rmps, &status->n_rmps);
1834 /* Spanning Tree. */
1837 send_bpdu_cb(struct ofpbuf *pkt, int port_num, void *ofproto_)
1839 struct ofproto_dpif *ofproto = ofproto_;
1840 struct stp_port *sp = stp_get_port(ofproto->stp, port_num);
1841 struct ofport_dpif *ofport;
1843 ofport = stp_port_get_aux(sp);
1845 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on unknown port %d",
1846 ofproto->up.name, port_num);
1848 struct eth_header *eth = pkt->l2;
1850 netdev_get_etheraddr(ofport->up.netdev, eth->eth_src);
1851 if (eth_addr_is_zero(eth->eth_src)) {
1852 VLOG_WARN_RL(&rl, "%s: cannot send BPDU on port %d "
1853 "with unknown MAC", ofproto->up.name, port_num);
1855 send_packet(ofport, pkt);
1861 /* Configures STP on 'ofproto_' using the settings defined in 's'. */
1863 set_stp(struct ofproto *ofproto_, const struct ofproto_stp_settings *s)
1865 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1867 /* Only revalidate flows if the configuration changed. */
1868 if (!s != !ofproto->stp) {
1869 ofproto->backer->need_revalidate = REV_RECONFIGURE;
1873 if (!ofproto->stp) {
1874 ofproto->stp = stp_create(ofproto_->name, s->system_id,
1875 send_bpdu_cb, ofproto);
1876 ofproto->stp_last_tick = time_msec();
1879 stp_set_bridge_id(ofproto->stp, s->system_id);
1880 stp_set_bridge_priority(ofproto->stp, s->priority);
1881 stp_set_hello_time(ofproto->stp, s->hello_time);
1882 stp_set_max_age(ofproto->stp, s->max_age);
1883 stp_set_forward_delay(ofproto->stp, s->fwd_delay);
1885 struct ofport *ofport;
1887 HMAP_FOR_EACH (ofport, hmap_node, &ofproto->up.ports) {
1888 set_stp_port(ofport, NULL);
1891 stp_destroy(ofproto->stp);
1892 ofproto->stp = NULL;
1899 get_stp_status(struct ofproto *ofproto_, struct ofproto_stp_status *s)
1901 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
1905 s->bridge_id = stp_get_bridge_id(ofproto->stp);
1906 s->designated_root = stp_get_designated_root(ofproto->stp);
1907 s->root_path_cost = stp_get_root_path_cost(ofproto->stp);
1916 update_stp_port_state(struct ofport_dpif *ofport)
1918 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1919 enum stp_state state;
1921 /* Figure out new state. */
1922 state = ofport->stp_port ? stp_port_get_state(ofport->stp_port)
1926 if (ofport->stp_state != state) {
1927 enum ofputil_port_state of_state;
1930 VLOG_DBG_RL(&rl, "port %s: STP state changed from %s to %s",
1931 netdev_get_name(ofport->up.netdev),
1932 stp_state_name(ofport->stp_state),
1933 stp_state_name(state));
1934 if (stp_learn_in_state(ofport->stp_state)
1935 != stp_learn_in_state(state)) {
1936 /* xxx Learning action flows should also be flushed. */
1937 mac_learning_flush(ofproto->ml,
1938 &ofproto->backer->revalidate_set);
1940 fwd_change = stp_forward_in_state(ofport->stp_state)
1941 != stp_forward_in_state(state);
1943 ofproto->backer->need_revalidate = REV_STP;
1944 ofport->stp_state = state;
1945 ofport->stp_state_entered = time_msec();
1947 if (fwd_change && ofport->bundle) {
1948 bundle_update(ofport->bundle);
1951 /* Update the STP state bits in the OpenFlow port description. */
1952 of_state = ofport->up.pp.state & ~OFPUTIL_PS_STP_MASK;
1953 of_state |= (state == STP_LISTENING ? OFPUTIL_PS_STP_LISTEN
1954 : state == STP_LEARNING ? OFPUTIL_PS_STP_LEARN
1955 : state == STP_FORWARDING ? OFPUTIL_PS_STP_FORWARD
1956 : state == STP_BLOCKING ? OFPUTIL_PS_STP_BLOCK
1958 ofproto_port_set_state(&ofport->up, of_state);
1962 /* Configures STP on 'ofport_' using the settings defined in 's'. The
1963 * caller is responsible for assigning STP port numbers and ensuring
1964 * there are no duplicates. */
1966 set_stp_port(struct ofport *ofport_,
1967 const struct ofproto_port_stp_settings *s)
1969 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
1970 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
1971 struct stp_port *sp = ofport->stp_port;
1973 if (!s || !s->enable) {
1975 ofport->stp_port = NULL;
1976 stp_port_disable(sp);
1977 update_stp_port_state(ofport);
1980 } else if (sp && stp_port_no(sp) != s->port_num
1981 && ofport == stp_port_get_aux(sp)) {
1982 /* The port-id changed, so disable the old one if it's not
1983 * already in use by another port. */
1984 stp_port_disable(sp);
1987 sp = ofport->stp_port = stp_get_port(ofproto->stp, s->port_num);
1988 stp_port_enable(sp);
1990 stp_port_set_aux(sp, ofport);
1991 stp_port_set_priority(sp, s->priority);
1992 stp_port_set_path_cost(sp, s->path_cost);
1994 update_stp_port_state(ofport);
2000 get_stp_port_status(struct ofport *ofport_,
2001 struct ofproto_port_stp_status *s)
2003 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2004 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2005 struct stp_port *sp = ofport->stp_port;
2007 if (!ofproto->stp || !sp) {
2013 s->port_id = stp_port_get_id(sp);
2014 s->state = stp_port_get_state(sp);
2015 s->sec_in_state = (time_msec() - ofport->stp_state_entered) / 1000;
2016 s->role = stp_port_get_role(sp);
2017 stp_port_get_counts(sp, &s->tx_count, &s->rx_count, &s->error_count);
2023 stp_run(struct ofproto_dpif *ofproto)
2026 long long int now = time_msec();
2027 long long int elapsed = now - ofproto->stp_last_tick;
2028 struct stp_port *sp;
2031 stp_tick(ofproto->stp, MIN(INT_MAX, elapsed));
2032 ofproto->stp_last_tick = now;
2034 while (stp_get_changed_port(ofproto->stp, &sp)) {
2035 struct ofport_dpif *ofport = stp_port_get_aux(sp);
2038 update_stp_port_state(ofport);
2042 if (stp_check_and_reset_fdb_flush(ofproto->stp)) {
2043 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2049 stp_wait(struct ofproto_dpif *ofproto)
2052 poll_timer_wait(1000);
2056 /* Returns true if STP should process 'flow'. */
2058 stp_should_process_flow(const struct flow *flow)
2060 return eth_addr_equals(flow->dl_dst, eth_addr_stp);
2064 stp_process_packet(const struct ofport_dpif *ofport,
2065 const struct ofpbuf *packet)
2067 struct ofpbuf payload = *packet;
2068 struct eth_header *eth = payload.data;
2069 struct stp_port *sp = ofport->stp_port;
2071 /* Sink packets on ports that have STP disabled when the bridge has
2073 if (!sp || stp_port_get_state(sp) == STP_DISABLED) {
2077 /* Trim off padding on payload. */
2078 if (payload.size > ntohs(eth->eth_type) + ETH_HEADER_LEN) {
2079 payload.size = ntohs(eth->eth_type) + ETH_HEADER_LEN;
2082 if (ofpbuf_try_pull(&payload, ETH_HEADER_LEN + LLC_HEADER_LEN)) {
2083 stp_received_bpdu(sp, payload.data, payload.size);
2087 static struct priority_to_dscp *
2088 get_priority(const struct ofport_dpif *ofport, uint32_t priority)
2090 struct priority_to_dscp *pdscp;
2093 hash = hash_int(priority, 0);
2094 HMAP_FOR_EACH_IN_BUCKET (pdscp, hmap_node, hash, &ofport->priorities) {
2095 if (pdscp->priority == priority) {
2103 ofport_clear_priorities(struct ofport_dpif *ofport)
2105 struct priority_to_dscp *pdscp, *next;
2107 HMAP_FOR_EACH_SAFE (pdscp, next, hmap_node, &ofport->priorities) {
2108 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2114 set_queues(struct ofport *ofport_,
2115 const struct ofproto_port_queue *qdscp_list,
2118 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
2119 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
2120 struct hmap new = HMAP_INITIALIZER(&new);
2123 for (i = 0; i < n_qdscp; i++) {
2124 struct priority_to_dscp *pdscp;
2128 dscp = (qdscp_list[i].dscp << 2) & IP_DSCP_MASK;
2129 if (dpif_queue_to_priority(ofproto->backer->dpif, qdscp_list[i].queue,
2134 pdscp = get_priority(ofport, priority);
2136 hmap_remove(&ofport->priorities, &pdscp->hmap_node);
2138 pdscp = xmalloc(sizeof *pdscp);
2139 pdscp->priority = priority;
2141 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2144 if (pdscp->dscp != dscp) {
2146 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2149 hmap_insert(&new, &pdscp->hmap_node, hash_int(pdscp->priority, 0));
2152 if (!hmap_is_empty(&ofport->priorities)) {
2153 ofport_clear_priorities(ofport);
2154 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2157 hmap_swap(&new, &ofport->priorities);
2165 /* Expires all MAC learning entries associated with 'bundle' and forces its
2166 * ofproto to revalidate every flow.
2168 * Normally MAC learning entries are removed only from the ofproto associated
2169 * with 'bundle', but if 'all_ofprotos' is true, then the MAC learning entries
2170 * are removed from every ofproto. When patch ports and SLB bonds are in use
2171 * and a VM migration happens and the gratuitous ARPs are somehow lost, this
2172 * avoids a MAC_ENTRY_IDLE_TIME delay before the migrated VM can communicate
2173 * with the host from which it migrated. */
2175 bundle_flush_macs(struct ofbundle *bundle, bool all_ofprotos)
2177 struct ofproto_dpif *ofproto = bundle->ofproto;
2178 struct mac_learning *ml = ofproto->ml;
2179 struct mac_entry *mac, *next_mac;
2181 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2182 LIST_FOR_EACH_SAFE (mac, next_mac, lru_node, &ml->lrus) {
2183 if (mac->port.p == bundle) {
2185 struct ofproto_dpif *o;
2187 HMAP_FOR_EACH (o, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2189 struct mac_entry *e;
2191 e = mac_learning_lookup(o->ml, mac->mac, mac->vlan,
2194 mac_learning_expire(o->ml, e);
2200 mac_learning_expire(ml, mac);
2205 static struct ofbundle *
2206 bundle_lookup(const struct ofproto_dpif *ofproto, void *aux)
2208 struct ofbundle *bundle;
2210 HMAP_FOR_EACH_IN_BUCKET (bundle, hmap_node, hash_pointer(aux, 0),
2211 &ofproto->bundles) {
2212 if (bundle->aux == aux) {
2219 /* Looks up each of the 'n_auxes' pointers in 'auxes' as bundles and adds the
2220 * ones that are found to 'bundles'. */
2222 bundle_lookup_multiple(struct ofproto_dpif *ofproto,
2223 void **auxes, size_t n_auxes,
2224 struct hmapx *bundles)
2228 hmapx_init(bundles);
2229 for (i = 0; i < n_auxes; i++) {
2230 struct ofbundle *bundle = bundle_lookup(ofproto, auxes[i]);
2232 hmapx_add(bundles, bundle);
2238 bundle_update(struct ofbundle *bundle)
2240 struct ofport_dpif *port;
2242 bundle->floodable = true;
2243 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2244 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2245 || !stp_forward_in_state(port->stp_state)) {
2246 bundle->floodable = false;
2253 bundle_del_port(struct ofport_dpif *port)
2255 struct ofbundle *bundle = port->bundle;
2257 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2259 list_remove(&port->bundle_node);
2260 port->bundle = NULL;
2263 lacp_slave_unregister(bundle->lacp, port);
2266 bond_slave_unregister(bundle->bond, port);
2269 bundle_update(bundle);
2273 bundle_add_port(struct ofbundle *bundle, uint32_t ofp_port,
2274 struct lacp_slave_settings *lacp)
2276 struct ofport_dpif *port;
2278 port = get_ofp_port(bundle->ofproto, ofp_port);
2283 if (port->bundle != bundle) {
2284 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2286 bundle_del_port(port);
2289 port->bundle = bundle;
2290 list_push_back(&bundle->ports, &port->bundle_node);
2291 if (port->up.pp.config & OFPUTIL_PC_NO_FLOOD
2292 || !stp_forward_in_state(port->stp_state)) {
2293 bundle->floodable = false;
2297 bundle->ofproto->backer->need_revalidate = REV_RECONFIGURE;
2298 lacp_slave_register(bundle->lacp, port, lacp);
2305 bundle_destroy(struct ofbundle *bundle)
2307 struct ofproto_dpif *ofproto;
2308 struct ofport_dpif *port, *next_port;
2315 ofproto = bundle->ofproto;
2316 for (i = 0; i < MAX_MIRRORS; i++) {
2317 struct ofmirror *m = ofproto->mirrors[i];
2319 if (m->out == bundle) {
2321 } else if (hmapx_find_and_delete(&m->srcs, bundle)
2322 || hmapx_find_and_delete(&m->dsts, bundle)) {
2323 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2328 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2329 bundle_del_port(port);
2332 bundle_flush_macs(bundle, true);
2333 hmap_remove(&ofproto->bundles, &bundle->hmap_node);
2335 free(bundle->trunks);
2336 lacp_destroy(bundle->lacp);
2337 bond_destroy(bundle->bond);
2342 bundle_set(struct ofproto *ofproto_, void *aux,
2343 const struct ofproto_bundle_settings *s)
2345 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2346 bool need_flush = false;
2347 struct ofport_dpif *port;
2348 struct ofbundle *bundle;
2349 unsigned long *trunks;
2355 bundle_destroy(bundle_lookup(ofproto, aux));
2359 ovs_assert(s->n_slaves == 1 || s->bond != NULL);
2360 ovs_assert((s->lacp != NULL) == (s->lacp_slaves != NULL));
2362 bundle = bundle_lookup(ofproto, aux);
2364 bundle = xmalloc(sizeof *bundle);
2366 bundle->ofproto = ofproto;
2367 hmap_insert(&ofproto->bundles, &bundle->hmap_node,
2368 hash_pointer(aux, 0));
2370 bundle->name = NULL;
2372 list_init(&bundle->ports);
2373 bundle->vlan_mode = PORT_VLAN_TRUNK;
2375 bundle->trunks = NULL;
2376 bundle->use_priority_tags = s->use_priority_tags;
2377 bundle->lacp = NULL;
2378 bundle->bond = NULL;
2380 bundle->floodable = true;
2382 bundle->src_mirrors = 0;
2383 bundle->dst_mirrors = 0;
2384 bundle->mirror_out = 0;
2387 if (!bundle->name || strcmp(s->name, bundle->name)) {
2389 bundle->name = xstrdup(s->name);
2394 if (!bundle->lacp) {
2395 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2396 bundle->lacp = lacp_create();
2398 lacp_configure(bundle->lacp, s->lacp);
2400 lacp_destroy(bundle->lacp);
2401 bundle->lacp = NULL;
2404 /* Update set of ports. */
2406 for (i = 0; i < s->n_slaves; i++) {
2407 if (!bundle_add_port(bundle, s->slaves[i],
2408 s->lacp ? &s->lacp_slaves[i] : NULL)) {
2412 if (!ok || list_size(&bundle->ports) != s->n_slaves) {
2413 struct ofport_dpif *next_port;
2415 LIST_FOR_EACH_SAFE (port, next_port, bundle_node, &bundle->ports) {
2416 for (i = 0; i < s->n_slaves; i++) {
2417 if (s->slaves[i] == port->up.ofp_port) {
2422 bundle_del_port(port);
2426 ovs_assert(list_size(&bundle->ports) <= s->n_slaves);
2428 if (list_is_empty(&bundle->ports)) {
2429 bundle_destroy(bundle);
2433 /* Set VLAN tagging mode */
2434 if (s->vlan_mode != bundle->vlan_mode
2435 || s->use_priority_tags != bundle->use_priority_tags) {
2436 bundle->vlan_mode = s->vlan_mode;
2437 bundle->use_priority_tags = s->use_priority_tags;
2442 vlan = (s->vlan_mode == PORT_VLAN_TRUNK ? -1
2443 : s->vlan >= 0 && s->vlan <= 4095 ? s->vlan
2445 if (vlan != bundle->vlan) {
2446 bundle->vlan = vlan;
2450 /* Get trunked VLANs. */
2451 switch (s->vlan_mode) {
2452 case PORT_VLAN_ACCESS:
2456 case PORT_VLAN_TRUNK:
2457 trunks = CONST_CAST(unsigned long *, s->trunks);
2460 case PORT_VLAN_NATIVE_UNTAGGED:
2461 case PORT_VLAN_NATIVE_TAGGED:
2462 if (vlan != 0 && (!s->trunks
2463 || !bitmap_is_set(s->trunks, vlan)
2464 || bitmap_is_set(s->trunks, 0))) {
2465 /* Force trunking the native VLAN and prohibit trunking VLAN 0. */
2467 trunks = bitmap_clone(s->trunks, 4096);
2469 trunks = bitmap_allocate1(4096);
2471 bitmap_set1(trunks, vlan);
2472 bitmap_set0(trunks, 0);
2474 trunks = CONST_CAST(unsigned long *, s->trunks);
2481 if (!vlan_bitmap_equal(trunks, bundle->trunks)) {
2482 free(bundle->trunks);
2483 if (trunks == s->trunks) {
2484 bundle->trunks = vlan_bitmap_clone(trunks);
2486 bundle->trunks = trunks;
2491 if (trunks != s->trunks) {
2496 if (!list_is_short(&bundle->ports)) {
2497 bundle->ofproto->has_bonded_bundles = true;
2499 if (bond_reconfigure(bundle->bond, s->bond)) {
2500 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2503 bundle->bond = bond_create(s->bond);
2504 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2507 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2508 bond_slave_register(bundle->bond, port, port->up.netdev);
2511 bond_destroy(bundle->bond);
2512 bundle->bond = NULL;
2515 /* If we changed something that would affect MAC learning, un-learn
2516 * everything on this port and force flow revalidation. */
2518 bundle_flush_macs(bundle, false);
2525 bundle_remove(struct ofport *port_)
2527 struct ofport_dpif *port = ofport_dpif_cast(port_);
2528 struct ofbundle *bundle = port->bundle;
2531 bundle_del_port(port);
2532 if (list_is_empty(&bundle->ports)) {
2533 bundle_destroy(bundle);
2534 } else if (list_is_short(&bundle->ports)) {
2535 bond_destroy(bundle->bond);
2536 bundle->bond = NULL;
2542 send_pdu_cb(void *port_, const void *pdu, size_t pdu_size)
2544 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 10);
2545 struct ofport_dpif *port = port_;
2546 uint8_t ea[ETH_ADDR_LEN];
2549 error = netdev_get_etheraddr(port->up.netdev, ea);
2551 struct ofpbuf packet;
2554 ofpbuf_init(&packet, 0);
2555 packet_pdu = eth_compose(&packet, eth_addr_lacp, ea, ETH_TYPE_LACP,
2557 memcpy(packet_pdu, pdu, pdu_size);
2559 send_packet(port, &packet);
2560 ofpbuf_uninit(&packet);
2562 VLOG_ERR_RL(&rl, "port %s: cannot obtain Ethernet address of iface "
2563 "%s (%s)", port->bundle->name,
2564 netdev_get_name(port->up.netdev), strerror(error));
2569 bundle_send_learning_packets(struct ofbundle *bundle)
2571 struct ofproto_dpif *ofproto = bundle->ofproto;
2572 int error, n_packets, n_errors;
2573 struct mac_entry *e;
2575 error = n_packets = n_errors = 0;
2576 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
2577 if (e->port.p != bundle) {
2578 struct ofpbuf *learning_packet;
2579 struct ofport_dpif *port;
2583 /* The assignment to "port" is unnecessary but makes "grep"ing for
2584 * struct ofport_dpif more effective. */
2585 learning_packet = bond_compose_learning_packet(bundle->bond,
2589 ret = send_packet(port, learning_packet);
2590 ofpbuf_delete(learning_packet);
2600 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
2601 VLOG_WARN_RL(&rl, "bond %s: %d errors sending %d gratuitous learning "
2602 "packets, last error was: %s",
2603 bundle->name, n_errors, n_packets, strerror(error));
2605 VLOG_DBG("bond %s: sent %d gratuitous learning packets",
2606 bundle->name, n_packets);
2611 bundle_run(struct ofbundle *bundle)
2614 lacp_run(bundle->lacp, send_pdu_cb);
2617 struct ofport_dpif *port;
2619 LIST_FOR_EACH (port, bundle_node, &bundle->ports) {
2620 bond_slave_set_may_enable(bundle->bond, port, port->may_enable);
2623 bond_run(bundle->bond, &bundle->ofproto->backer->revalidate_set,
2624 lacp_status(bundle->lacp));
2625 if (bond_should_send_learning_packets(bundle->bond)) {
2626 bundle_send_learning_packets(bundle);
2632 bundle_wait(struct ofbundle *bundle)
2635 lacp_wait(bundle->lacp);
2638 bond_wait(bundle->bond);
2645 mirror_scan(struct ofproto_dpif *ofproto)
2649 for (idx = 0; idx < MAX_MIRRORS; idx++) {
2650 if (!ofproto->mirrors[idx]) {
2657 static struct ofmirror *
2658 mirror_lookup(struct ofproto_dpif *ofproto, void *aux)
2662 for (i = 0; i < MAX_MIRRORS; i++) {
2663 struct ofmirror *mirror = ofproto->mirrors[i];
2664 if (mirror && mirror->aux == aux) {
2672 /* Update the 'dup_mirrors' member of each of the ofmirrors in 'ofproto'. */
2674 mirror_update_dups(struct ofproto_dpif *ofproto)
2678 for (i = 0; i < MAX_MIRRORS; i++) {
2679 struct ofmirror *m = ofproto->mirrors[i];
2682 m->dup_mirrors = MIRROR_MASK_C(1) << i;
2686 for (i = 0; i < MAX_MIRRORS; i++) {
2687 struct ofmirror *m1 = ofproto->mirrors[i];
2694 for (j = i + 1; j < MAX_MIRRORS; j++) {
2695 struct ofmirror *m2 = ofproto->mirrors[j];
2697 if (m2 && m1->out == m2->out && m1->out_vlan == m2->out_vlan) {
2698 m1->dup_mirrors |= MIRROR_MASK_C(1) << j;
2699 m2->dup_mirrors |= m1->dup_mirrors;
2706 mirror_set(struct ofproto *ofproto_, void *aux,
2707 const struct ofproto_mirror_settings *s)
2709 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2710 mirror_mask_t mirror_bit;
2711 struct ofbundle *bundle;
2712 struct ofmirror *mirror;
2713 struct ofbundle *out;
2714 struct hmapx srcs; /* Contains "struct ofbundle *"s. */
2715 struct hmapx dsts; /* Contains "struct ofbundle *"s. */
2718 mirror = mirror_lookup(ofproto, aux);
2720 mirror_destroy(mirror);
2726 idx = mirror_scan(ofproto);
2728 VLOG_WARN("bridge %s: maximum of %d port mirrors reached, "
2730 ofproto->up.name, MAX_MIRRORS, s->name);
2734 mirror = ofproto->mirrors[idx] = xzalloc(sizeof *mirror);
2735 mirror->ofproto = ofproto;
2738 mirror->out_vlan = -1;
2739 mirror->name = NULL;
2742 if (!mirror->name || strcmp(s->name, mirror->name)) {
2744 mirror->name = xstrdup(s->name);
2747 /* Get the new configuration. */
2748 if (s->out_bundle) {
2749 out = bundle_lookup(ofproto, s->out_bundle);
2751 mirror_destroy(mirror);
2757 out_vlan = s->out_vlan;
2759 bundle_lookup_multiple(ofproto, s->srcs, s->n_srcs, &srcs);
2760 bundle_lookup_multiple(ofproto, s->dsts, s->n_dsts, &dsts);
2762 /* If the configuration has not changed, do nothing. */
2763 if (hmapx_equals(&srcs, &mirror->srcs)
2764 && hmapx_equals(&dsts, &mirror->dsts)
2765 && vlan_bitmap_equal(mirror->vlans, s->src_vlans)
2766 && mirror->out == out
2767 && mirror->out_vlan == out_vlan)
2769 hmapx_destroy(&srcs);
2770 hmapx_destroy(&dsts);
2774 hmapx_swap(&srcs, &mirror->srcs);
2775 hmapx_destroy(&srcs);
2777 hmapx_swap(&dsts, &mirror->dsts);
2778 hmapx_destroy(&dsts);
2780 free(mirror->vlans);
2781 mirror->vlans = vlan_bitmap_clone(s->src_vlans);
2784 mirror->out_vlan = out_vlan;
2786 /* Update bundles. */
2787 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2788 HMAP_FOR_EACH (bundle, hmap_node, &mirror->ofproto->bundles) {
2789 if (hmapx_contains(&mirror->srcs, bundle)) {
2790 bundle->src_mirrors |= mirror_bit;
2792 bundle->src_mirrors &= ~mirror_bit;
2795 if (hmapx_contains(&mirror->dsts, bundle)) {
2796 bundle->dst_mirrors |= mirror_bit;
2798 bundle->dst_mirrors &= ~mirror_bit;
2801 if (mirror->out == bundle) {
2802 bundle->mirror_out |= mirror_bit;
2804 bundle->mirror_out &= ~mirror_bit;
2808 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2809 ofproto->has_mirrors = true;
2810 mac_learning_flush(ofproto->ml,
2811 &ofproto->backer->revalidate_set);
2812 mirror_update_dups(ofproto);
2818 mirror_destroy(struct ofmirror *mirror)
2820 struct ofproto_dpif *ofproto;
2821 mirror_mask_t mirror_bit;
2822 struct ofbundle *bundle;
2829 ofproto = mirror->ofproto;
2830 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2831 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2833 mirror_bit = MIRROR_MASK_C(1) << mirror->idx;
2834 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
2835 bundle->src_mirrors &= ~mirror_bit;
2836 bundle->dst_mirrors &= ~mirror_bit;
2837 bundle->mirror_out &= ~mirror_bit;
2840 hmapx_destroy(&mirror->srcs);
2841 hmapx_destroy(&mirror->dsts);
2842 free(mirror->vlans);
2844 ofproto->mirrors[mirror->idx] = NULL;
2848 mirror_update_dups(ofproto);
2850 ofproto->has_mirrors = false;
2851 for (i = 0; i < MAX_MIRRORS; i++) {
2852 if (ofproto->mirrors[i]) {
2853 ofproto->has_mirrors = true;
2860 mirror_get_stats(struct ofproto *ofproto_, void *aux,
2861 uint64_t *packets, uint64_t *bytes)
2863 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2864 struct ofmirror *mirror = mirror_lookup(ofproto, aux);
2867 *packets = *bytes = UINT64_MAX;
2871 *packets = mirror->packet_count;
2872 *bytes = mirror->byte_count;
2878 set_flood_vlans(struct ofproto *ofproto_, unsigned long *flood_vlans)
2880 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2881 if (mac_learning_set_flood_vlans(ofproto->ml, flood_vlans)) {
2882 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
2888 is_mirror_output_bundle(const struct ofproto *ofproto_, void *aux)
2890 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2891 struct ofbundle *bundle = bundle_lookup(ofproto, aux);
2892 return bundle && bundle->mirror_out != 0;
2896 forward_bpdu_changed(struct ofproto *ofproto_)
2898 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2899 ofproto->backer->need_revalidate = REV_RECONFIGURE;
2903 set_mac_table_config(struct ofproto *ofproto_, unsigned int idle_time,
2906 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
2907 mac_learning_set_idle_time(ofproto->ml, idle_time);
2908 mac_learning_set_max_entries(ofproto->ml, max_entries);
2913 static struct ofport_dpif *
2914 get_ofp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
2916 struct ofport *ofport = ofproto_get_port(&ofproto->up, ofp_port);
2917 return ofport ? ofport_dpif_cast(ofport) : NULL;
2920 static struct ofport_dpif *
2921 get_odp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
2923 struct ofport_dpif *port = odp_port_to_ofport(ofproto->backer, odp_port);
2924 return port && &ofproto->up == port->up.ofproto ? port : NULL;
2928 ofproto_port_from_dpif_port(struct ofproto_dpif *ofproto,
2929 struct ofproto_port *ofproto_port,
2930 struct dpif_port *dpif_port)
2932 ofproto_port->name = dpif_port->name;
2933 ofproto_port->type = dpif_port->type;
2934 ofproto_port->ofp_port = odp_port_to_ofp_port(ofproto, dpif_port->port_no);
2937 static struct ofport_dpif *
2938 ofport_get_peer(const struct ofport_dpif *ofport_dpif)
2940 const struct ofproto_dpif *ofproto;
2943 peer = netdev_vport_patch_peer(ofport_dpif->up.netdev);
2948 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
2949 struct ofport *ofport;
2951 ofport = shash_find_data(&ofproto->up.port_by_name, peer);
2952 if (ofport && ofport->ofproto->ofproto_class == &ofproto_dpif_class) {
2953 return ofport_dpif_cast(ofport);
2960 port_run_fast(struct ofport_dpif *ofport)
2962 if (ofport->cfm && cfm_should_send_ccm(ofport->cfm)) {
2963 struct ofpbuf packet;
2965 ofpbuf_init(&packet, 0);
2966 cfm_compose_ccm(ofport->cfm, &packet, ofport->up.pp.hw_addr);
2967 send_packet(ofport, &packet);
2968 ofpbuf_uninit(&packet);
2973 port_run(struct ofport_dpif *ofport)
2975 long long int carrier_seq = netdev_get_carrier_resets(ofport->up.netdev);
2976 bool carrier_changed = carrier_seq != ofport->carrier_seq;
2977 bool enable = netdev_get_carrier(ofport->up.netdev);
2979 ofport->carrier_seq = carrier_seq;
2981 port_run_fast(ofport);
2983 if (ofport->tnl_port
2984 && tnl_port_reconfigure(&ofport->up, ofport->odp_port,
2985 &ofport->tnl_port)) {
2986 ofproto_dpif_cast(ofport->up.ofproto)->backer->need_revalidate = true;
2990 int cfm_opup = cfm_get_opup(ofport->cfm);
2992 cfm_run(ofport->cfm);
2993 enable = enable && !cfm_get_fault(ofport->cfm);
2995 if (cfm_opup >= 0) {
2996 enable = enable && cfm_opup;
3000 if (ofport->bundle) {
3001 enable = enable && lacp_slave_may_enable(ofport->bundle->lacp, ofport);
3002 if (carrier_changed) {
3003 lacp_slave_carrier_changed(ofport->bundle->lacp, ofport);
3007 if (ofport->may_enable != enable) {
3008 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3010 if (ofproto->has_bundle_action) {
3011 ofproto->backer->need_revalidate = REV_PORT_TOGGLED;
3015 ofport->may_enable = enable;
3019 port_wait(struct ofport_dpif *ofport)
3022 cfm_wait(ofport->cfm);
3027 port_query_by_name(const struct ofproto *ofproto_, const char *devname,
3028 struct ofproto_port *ofproto_port)
3030 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3031 struct dpif_port dpif_port;
3034 if (sset_contains(&ofproto->ghost_ports, devname)) {
3035 const char *type = netdev_get_type_from_name(devname);
3037 /* We may be called before ofproto->up.port_by_name is populated with
3038 * the appropriate ofport. For this reason, we must get the name and
3039 * type from the netdev layer directly. */
3041 const struct ofport *ofport;
3043 ofport = shash_find_data(&ofproto->up.port_by_name, devname);
3044 ofproto_port->ofp_port = ofport ? ofport->ofp_port : OFPP_NONE;
3045 ofproto_port->name = xstrdup(devname);
3046 ofproto_port->type = xstrdup(type);
3052 if (!sset_contains(&ofproto->ports, devname)) {
3055 error = dpif_port_query_by_name(ofproto->backer->dpif,
3056 devname, &dpif_port);
3058 ofproto_port_from_dpif_port(ofproto, ofproto_port, &dpif_port);
3064 port_add(struct ofproto *ofproto_, struct netdev *netdev)
3066 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3067 const char *dp_port_name = netdev_vport_get_dpif_port(netdev);
3068 const char *devname = netdev_get_name(netdev);
3070 if (netdev_vport_is_patch(netdev)) {
3071 sset_add(&ofproto->ghost_ports, netdev_get_name(netdev));
3075 if (!dpif_port_exists(ofproto->backer->dpif, dp_port_name)) {
3076 uint32_t port_no = UINT32_MAX;
3079 error = dpif_port_add(ofproto->backer->dpif, netdev, &port_no);
3083 if (netdev_get_tunnel_config(netdev)) {
3084 simap_put(&ofproto->backer->tnl_backers, dp_port_name, port_no);
3088 if (netdev_get_tunnel_config(netdev)) {
3089 sset_add(&ofproto->ghost_ports, devname);
3091 sset_add(&ofproto->ports, devname);
3097 port_del(struct ofproto *ofproto_, uint16_t ofp_port)
3099 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3100 struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
3107 sset_find_and_delete(&ofproto->ghost_ports,
3108 netdev_get_name(ofport->up.netdev));
3109 ofproto->backer->need_revalidate = REV_RECONFIGURE;
3110 if (!ofport->tnl_port) {
3111 error = dpif_port_del(ofproto->backer->dpif, ofport->odp_port);
3113 /* The caller is going to close ofport->up.netdev. If this is a
3114 * bonded port, then the bond is using that netdev, so remove it
3115 * from the bond. The client will need to reconfigure everything
3116 * after deleting ports, so then the slave will get re-added. */
3117 bundle_remove(&ofport->up);
3124 port_get_stats(const struct ofport *ofport_, struct netdev_stats *stats)
3126 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3129 error = netdev_get_stats(ofport->up.netdev, stats);
3131 if (!error && ofport_->ofp_port == OFPP_LOCAL) {
3132 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
3134 /* ofproto->stats.tx_packets represents packets that we created
3135 * internally and sent to some port (e.g. packets sent with
3136 * send_packet()). Account for them as if they had come from
3137 * OFPP_LOCAL and got forwarded. */
3139 if (stats->rx_packets != UINT64_MAX) {
3140 stats->rx_packets += ofproto->stats.tx_packets;
3143 if (stats->rx_bytes != UINT64_MAX) {
3144 stats->rx_bytes += ofproto->stats.tx_bytes;
3147 /* ofproto->stats.rx_packets represents packets that were received on
3148 * some port and we processed internally and dropped (e.g. STP).
3149 * Account for them as if they had been forwarded to OFPP_LOCAL. */
3151 if (stats->tx_packets != UINT64_MAX) {
3152 stats->tx_packets += ofproto->stats.rx_packets;
3155 if (stats->tx_bytes != UINT64_MAX) {
3156 stats->tx_bytes += ofproto->stats.rx_bytes;
3163 /* Account packets for LOCAL port. */
3165 ofproto_update_local_port_stats(const struct ofproto *ofproto_,
3166 size_t tx_size, size_t rx_size)
3168 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3171 ofproto->stats.rx_packets++;
3172 ofproto->stats.rx_bytes += rx_size;
3175 ofproto->stats.tx_packets++;
3176 ofproto->stats.tx_bytes += tx_size;
3180 struct port_dump_state {
3185 struct ofproto_port port;
3190 port_dump_start(const struct ofproto *ofproto_ OVS_UNUSED, void **statep)
3192 *statep = xzalloc(sizeof(struct port_dump_state));
3197 port_dump_next(const struct ofproto *ofproto_, void *state_,
3198 struct ofproto_port *port)
3200 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3201 struct port_dump_state *state = state_;
3202 const struct sset *sset;
3203 struct sset_node *node;
3205 if (state->has_port) {
3206 ofproto_port_destroy(&state->port);
3207 state->has_port = false;
3209 sset = state->ghost ? &ofproto->ghost_ports : &ofproto->ports;
3210 while ((node = sset_at_position(sset, &state->bucket, &state->offset))) {
3213 error = port_query_by_name(ofproto_, node->name, &state->port);
3215 *port = state->port;
3216 state->has_port = true;
3218 } else if (error != ENODEV) {
3223 if (!state->ghost) {
3224 state->ghost = true;
3227 return port_dump_next(ofproto_, state_, port);
3234 port_dump_done(const struct ofproto *ofproto_ OVS_UNUSED, void *state_)
3236 struct port_dump_state *state = state_;
3238 if (state->has_port) {
3239 ofproto_port_destroy(&state->port);
3246 port_poll(const struct ofproto *ofproto_, char **devnamep)
3248 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3250 if (ofproto->port_poll_errno) {
3251 int error = ofproto->port_poll_errno;
3252 ofproto->port_poll_errno = 0;
3256 if (sset_is_empty(&ofproto->port_poll_set)) {
3260 *devnamep = sset_pop(&ofproto->port_poll_set);
3265 port_poll_wait(const struct ofproto *ofproto_)
3267 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
3268 dpif_port_poll_wait(ofproto->backer->dpif);
3272 port_is_lacp_current(const struct ofport *ofport_)
3274 const struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
3275 return (ofport->bundle && ofport->bundle->lacp
3276 ? lacp_slave_is_current(ofport->bundle->lacp, ofport)
3280 /* Upcall handling. */
3282 /* Flow miss batching.
3284 * Some dpifs implement operations faster when you hand them off in a batch.
3285 * To allow batching, "struct flow_miss" queues the dpif-related work needed
3286 * for a given flow. Each "struct flow_miss" corresponds to sending one or
3287 * more packets, plus possibly installing the flow in the dpif.
3289 * So far we only batch the operations that affect flow setup time the most.
3290 * It's possible to batch more than that, but the benefit might be minimal. */
3292 struct hmap_node hmap_node;
3293 struct ofproto_dpif *ofproto;
3295 enum odp_key_fitness key_fitness;
3296 const struct nlattr *key;
3298 struct initial_vals initial_vals;
3299 struct list packets;
3300 enum dpif_upcall_type upcall_type;
3301 uint32_t odp_in_port;
3304 struct flow_miss_op {
3305 struct dpif_op dpif_op;
3306 void *garbage; /* Pointer to pass to free(), NULL if none. */
3307 uint64_t stub[1024 / 8]; /* Temporary buffer. */
3310 /* Sends an OFPT_PACKET_IN message for 'packet' of type OFPR_NO_MATCH to each
3311 * OpenFlow controller as necessary according to their individual
3312 * configurations. */
3314 send_packet_in_miss(struct ofproto_dpif *ofproto, const struct ofpbuf *packet,
3315 const struct flow *flow)
3317 struct ofputil_packet_in pin;
3319 pin.packet = packet->data;
3320 pin.packet_len = packet->size;
3321 pin.reason = OFPR_NO_MATCH;
3322 pin.controller_id = 0;
3327 pin.send_len = 0; /* not used for flow table misses */
3329 flow_get_metadata(flow, &pin.fmd);
3331 connmgr_send_packet_in(ofproto->up.connmgr, &pin);
3334 static enum slow_path_reason
3335 process_special(struct ofproto_dpif *ofproto, const struct flow *flow,
3336 const struct ofport_dpif *ofport, const struct ofpbuf *packet)
3340 } else if (ofport->cfm && cfm_should_process_flow(ofport->cfm, flow)) {
3342 cfm_process_heartbeat(ofport->cfm, packet);
3345 } else if (ofport->bundle && ofport->bundle->lacp
3346 && flow->dl_type == htons(ETH_TYPE_LACP)) {
3348 lacp_process_packet(ofport->bundle->lacp, ofport, packet);
3351 } else if (ofproto->stp && stp_should_process_flow(flow)) {
3353 stp_process_packet(ofport, packet);
3361 static struct flow_miss *
3362 flow_miss_find(struct hmap *todo, const struct ofproto_dpif *ofproto,
3363 const struct flow *flow, uint32_t hash)
3365 struct flow_miss *miss;
3367 HMAP_FOR_EACH_WITH_HASH (miss, hmap_node, hash, todo) {
3368 if (miss->ofproto == ofproto && flow_equal(&miss->flow, flow)) {
3376 /* Partially Initializes 'op' as an "execute" operation for 'miss' and
3377 * 'packet'. The caller must initialize op->actions and op->actions_len. If
3378 * 'miss' is associated with a subfacet the caller must also initialize the
3379 * returned op->subfacet, and if anything needs to be freed after processing
3380 * the op, the caller must initialize op->garbage also. */
3382 init_flow_miss_execute_op(struct flow_miss *miss, struct ofpbuf *packet,
3383 struct flow_miss_op *op)
3385 if (miss->flow.vlan_tci != miss->initial_vals.vlan_tci) {
3386 /* This packet was received on a VLAN splinter port. We
3387 * added a VLAN to the packet to make the packet resemble
3388 * the flow, but the actions were composed assuming that
3389 * the packet contained no VLAN. So, we must remove the
3390 * VLAN header from the packet before trying to execute the
3392 eth_pop_vlan(packet);
3396 op->dpif_op.type = DPIF_OP_EXECUTE;
3397 op->dpif_op.u.execute.key = miss->key;
3398 op->dpif_op.u.execute.key_len = miss->key_len;
3399 op->dpif_op.u.execute.packet = packet;
3402 /* Helper for handle_flow_miss_without_facet() and
3403 * handle_flow_miss_with_facet(). */
3405 handle_flow_miss_common(struct rule_dpif *rule,
3406 struct ofpbuf *packet, const struct flow *flow)
3408 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3410 ofproto->n_matches++;
3412 if (rule->up.cr.priority == FAIL_OPEN_PRIORITY) {
3414 * Extra-special case for fail-open mode.
3416 * We are in fail-open mode and the packet matched the fail-open
3417 * rule, but we are connected to a controller too. We should send
3418 * the packet up to the controller in the hope that it will try to
3419 * set up a flow and thereby allow us to exit fail-open.
3421 * See the top-level comment in fail-open.c for more information.
3423 send_packet_in_miss(ofproto, packet, flow);
3427 /* Figures out whether a flow that missed in 'ofproto', whose details are in
3428 * 'miss', is likely to be worth tracking in detail in userspace and (usually)
3429 * installing a datapath flow. The answer is usually "yes" (a return value of
3430 * true). However, for short flows the cost of bookkeeping is much higher than
3431 * the benefits, so when the datapath holds a large number of flows we impose
3432 * some heuristics to decide which flows are likely to be worth tracking. */
3434 flow_miss_should_make_facet(struct ofproto_dpif *ofproto,
3435 struct flow_miss *miss, uint32_t hash)
3437 if (!ofproto->governor) {
3440 n_subfacets = hmap_count(&ofproto->subfacets);
3441 if (n_subfacets * 2 <= ofproto->up.flow_eviction_threshold) {
3445 ofproto->governor = governor_create(ofproto->up.name);
3448 return governor_should_install_flow(ofproto->governor, hash,
3449 list_size(&miss->packets));
3452 /* Handles 'miss', which matches 'rule', without creating a facet or subfacet
3453 * or creating any datapath flow. May add an "execute" operation to 'ops' and
3454 * increment '*n_ops'. */
3456 handle_flow_miss_without_facet(struct flow_miss *miss,
3457 struct rule_dpif *rule,
3458 struct flow_miss_op *ops, size_t *n_ops)
3460 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
3461 long long int now = time_msec();
3462 struct action_xlate_ctx ctx;
3463 struct ofpbuf *packet;
3465 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3466 struct flow_miss_op *op = &ops[*n_ops];
3467 struct dpif_flow_stats stats;
3468 struct ofpbuf odp_actions;
3470 COVERAGE_INC(facet_suppress);
3472 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3474 dpif_flow_stats_extract(&miss->flow, packet, now, &stats);
3475 rule_credit_stats(rule, &stats);
3477 action_xlate_ctx_init(&ctx, ofproto, &miss->flow,
3478 &miss->initial_vals, rule, 0, packet);
3479 ctx.resubmit_stats = &stats;
3480 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
3483 if (odp_actions.size) {
3484 struct dpif_execute *execute = &op->dpif_op.u.execute;
3486 init_flow_miss_execute_op(miss, packet, op);
3487 execute->actions = odp_actions.data;
3488 execute->actions_len = odp_actions.size;
3489 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3493 ofpbuf_uninit(&odp_actions);
3498 /* Handles 'miss', which matches 'facet'. May add any required datapath
3499 * operations to 'ops', incrementing '*n_ops' for each new op.
3501 * All of the packets in 'miss' are considered to have arrived at time 'now'.
3502 * This is really important only for new facets: if we just called time_msec()
3503 * here, then the new subfacet or its packets could look (occasionally) as
3504 * though it was used some time after the facet was used. That can make a
3505 * one-packet flow look like it has a nonzero duration, which looks odd in
3506 * e.g. NetFlow statistics. */
3508 handle_flow_miss_with_facet(struct flow_miss *miss, struct facet *facet,
3510 struct flow_miss_op *ops, size_t *n_ops)
3512 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
3513 enum subfacet_path want_path;
3514 struct subfacet *subfacet;
3515 struct ofpbuf *packet;
3517 subfacet = subfacet_create(facet, miss, now);
3519 LIST_FOR_EACH (packet, list_node, &miss->packets) {
3520 struct flow_miss_op *op = &ops[*n_ops];
3521 struct dpif_flow_stats stats;
3522 struct ofpbuf odp_actions;
3524 handle_flow_miss_common(facet->rule, packet, &miss->flow);
3526 ofpbuf_use_stub(&odp_actions, op->stub, sizeof op->stub);
3527 if (!subfacet->actions || subfacet->slow) {
3528 subfacet_make_actions(subfacet, packet, &odp_actions);
3531 dpif_flow_stats_extract(&facet->flow, packet, now, &stats);
3532 subfacet_update_stats(subfacet, &stats);
3534 if (subfacet->actions_len) {
3535 struct dpif_execute *execute = &op->dpif_op.u.execute;
3537 init_flow_miss_execute_op(miss, packet, op);
3538 if (!subfacet->slow) {
3539 execute->actions = subfacet->actions;
3540 execute->actions_len = subfacet->actions_len;
3541 ofpbuf_uninit(&odp_actions);
3543 execute->actions = odp_actions.data;
3544 execute->actions_len = odp_actions.size;
3545 op->garbage = ofpbuf_get_uninit_pointer(&odp_actions);
3550 ofpbuf_uninit(&odp_actions);
3554 want_path = subfacet_want_path(subfacet->slow);
3555 if (miss->upcall_type == DPIF_UC_MISS || subfacet->path != want_path) {
3556 struct flow_miss_op *op = &ops[(*n_ops)++];
3557 struct dpif_flow_put *put = &op->dpif_op.u.flow_put;
3559 subfacet->path = want_path;
3562 op->dpif_op.type = DPIF_OP_FLOW_PUT;
3563 put->flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
3564 put->key = miss->key;
3565 put->key_len = miss->key_len;
3566 if (want_path == SF_FAST_PATH) {
3567 put->actions = subfacet->actions;
3568 put->actions_len = subfacet->actions_len;
3570 compose_slow_path(ofproto, &facet->flow, subfacet->slow,
3571 op->stub, sizeof op->stub,
3572 &put->actions, &put->actions_len);
3578 /* Handles flow miss 'miss'. May add any required datapath operations
3579 * to 'ops', incrementing '*n_ops' for each new op. */
3581 handle_flow_miss(struct flow_miss *miss, struct flow_miss_op *ops,
3584 struct ofproto_dpif *ofproto = miss->ofproto;
3585 struct facet *facet;
3589 /* The caller must ensure that miss->hmap_node.hash contains
3590 * flow_hash(miss->flow, 0). */
3591 hash = miss->hmap_node.hash;
3593 facet = facet_lookup_valid(ofproto, &miss->flow, hash);
3595 struct rule_dpif *rule = rule_dpif_lookup(ofproto, &miss->flow);
3597 if (!flow_miss_should_make_facet(ofproto, miss, hash)) {
3598 handle_flow_miss_without_facet(miss, rule, ops, n_ops);
3602 facet = facet_create(rule, &miss->flow, hash);
3607 handle_flow_miss_with_facet(miss, facet, now, ops, n_ops);
3610 static struct drop_key *
3611 drop_key_lookup(const struct dpif_backer *backer, const struct nlattr *key,
3614 struct drop_key *drop_key;
3616 HMAP_FOR_EACH_WITH_HASH (drop_key, hmap_node, hash_bytes(key, key_len, 0),
3617 &backer->drop_keys) {
3618 if (drop_key->key_len == key_len
3619 && !memcmp(drop_key->key, key, key_len)) {
3627 drop_key_clear(struct dpif_backer *backer)
3629 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
3630 struct drop_key *drop_key, *next;
3632 HMAP_FOR_EACH_SAFE (drop_key, next, hmap_node, &backer->drop_keys) {
3635 error = dpif_flow_del(backer->dpif, drop_key->key, drop_key->key_len,
3637 if (error && !VLOG_DROP_WARN(&rl)) {
3638 struct ds ds = DS_EMPTY_INITIALIZER;
3639 odp_flow_key_format(drop_key->key, drop_key->key_len, &ds);
3640 VLOG_WARN("Failed to delete drop key (%s) (%s)", strerror(error),
3645 hmap_remove(&backer->drop_keys, &drop_key->hmap_node);
3646 free(drop_key->key);
3651 /* Given a datpath, packet, and flow metadata ('backer', 'packet', and 'key'
3652 * respectively), populates 'flow' with the result of odp_flow_key_to_flow().
3653 * Optionally, if nonnull, populates 'fitnessp' with the fitness of 'flow' as
3654 * returned by odp_flow_key_to_flow(). Also, optionally populates 'ofproto'
3655 * with the ofproto_dpif, and 'odp_in_port' with the datapath in_port, that
3656 * 'packet' ingressed.
3658 * If 'ofproto' is nonnull, requires 'flow''s in_port to exist. Otherwise sets
3659 * 'flow''s in_port to OFPP_NONE.
3661 * This function does post-processing on data returned from
3662 * odp_flow_key_to_flow() to help make VLAN splinters transparent to the rest
3663 * of the upcall processing logic. In particular, if the extracted in_port is
3664 * a VLAN splinter port, it replaces flow->in_port by the "real" port, sets
3665 * flow->vlan_tci correctly for the VLAN of the VLAN splinter port, and pushes
3666 * a VLAN header onto 'packet' (if it is nonnull).
3668 * Optionally, if 'initial_vals' is nonnull, sets 'initial_vals->vlan_tci'
3669 * to the VLAN TCI with which the packet was really received, that is, the
3670 * actual VLAN TCI extracted by odp_flow_key_to_flow(). (This differs from
3671 * the value returned in flow->vlan_tci only for packets received on
3672 * VLAN splinters.) Also, if received on an IP tunnel, sets
3673 * 'initial_vals->tunnel_ip_tos' to the tunnel's IP TOS.
3675 * Similarly, this function also includes some logic to help with tunnels. It
3676 * may modify 'flow' as necessary to make the tunneling implementation
3677 * transparent to the upcall processing logic.
3679 * Returns 0 if successful, ENODEV if the parsed flow has no associated ofport,
3680 * or some other positive errno if there are other problems. */
3682 ofproto_receive(const struct dpif_backer *backer, struct ofpbuf *packet,
3683 const struct nlattr *key, size_t key_len,
3684 struct flow *flow, enum odp_key_fitness *fitnessp,
3685 struct ofproto_dpif **ofproto, uint32_t *odp_in_port,
3686 struct initial_vals *initial_vals)
3688 const struct ofport_dpif *port;
3689 enum odp_key_fitness fitness;
3692 fitness = odp_flow_key_to_flow(key, key_len, flow);
3693 if (fitness == ODP_FIT_ERROR) {
3699 initial_vals->vlan_tci = flow->vlan_tci;
3700 initial_vals->tunnel_ip_tos = flow->tunnel.ip_tos;
3704 *odp_in_port = flow->in_port;
3707 if (tnl_port_should_receive(flow)) {
3708 const struct ofport *ofport = tnl_port_receive(flow);
3710 flow->in_port = OFPP_NONE;
3713 port = ofport_dpif_cast(ofport);
3715 /* We can't reproduce 'key' from 'flow'. */
3716 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3718 /* XXX: Since the tunnel module is not scoped per backer, it's
3719 * theoretically possible that we'll receive an ofport belonging to an
3720 * entirely different datapath. In practice, this can't happen because
3721 * no platforms has two separate datapaths which each support
3723 ovs_assert(ofproto_dpif_cast(port->up.ofproto)->backer == backer);
3725 port = odp_port_to_ofport(backer, flow->in_port);
3727 flow->in_port = OFPP_NONE;
3731 flow->in_port = port->up.ofp_port;
3732 if (vsp_adjust_flow(ofproto_dpif_cast(port->up.ofproto), flow)) {
3734 /* Make the packet resemble the flow, so that it gets sent to
3735 * an OpenFlow controller properly, so that it looks correct
3736 * for sFlow, and so that flow_extract() will get the correct
3737 * vlan_tci if it is called on 'packet'.
3739 * The allocated space inside 'packet' probably also contains
3740 * 'key', that is, both 'packet' and 'key' are probably part of
3741 * a struct dpif_upcall (see the large comment on that
3742 * structure definition), so pushing data on 'packet' is in
3743 * general not a good idea since it could overwrite 'key' or
3744 * free it as a side effect. However, it's OK in this special
3745 * case because we know that 'packet' is inside a Netlink
3746 * attribute: pushing 4 bytes will just overwrite the 4-byte
3747 * "struct nlattr", which is fine since we don't need that
3748 * header anymore. */
3749 eth_push_vlan(packet, flow->vlan_tci);
3751 /* We can't reproduce 'key' from 'flow'. */
3752 fitness = fitness == ODP_FIT_PERFECT ? ODP_FIT_TOO_MUCH : fitness;
3758 *ofproto = ofproto_dpif_cast(port->up.ofproto);
3763 *fitnessp = fitness;
3769 handle_miss_upcalls(struct dpif_backer *backer, struct dpif_upcall *upcalls,
3772 struct dpif_upcall *upcall;
3773 struct flow_miss *miss;
3774 struct flow_miss misses[FLOW_MISS_MAX_BATCH];
3775 struct flow_miss_op flow_miss_ops[FLOW_MISS_MAX_BATCH * 2];
3776 struct dpif_op *dpif_ops[FLOW_MISS_MAX_BATCH * 2];
3786 /* Construct the to-do list.
3788 * This just amounts to extracting the flow from each packet and sticking
3789 * the packets that have the same flow in the same "flow_miss" structure so
3790 * that we can process them together. */
3793 for (upcall = upcalls; upcall < &upcalls[n_upcalls]; upcall++) {
3794 struct flow_miss *miss = &misses[n_misses];
3795 struct flow_miss *existing_miss;
3796 struct ofproto_dpif *ofproto;
3797 uint32_t odp_in_port;
3802 error = ofproto_receive(backer, upcall->packet, upcall->key,
3803 upcall->key_len, &flow, &miss->key_fitness,
3804 &ofproto, &odp_in_port, &miss->initial_vals);
3805 if (error == ENODEV) {
3806 struct drop_key *drop_key;
3808 /* Received packet on port for which we couldn't associate
3809 * an ofproto. This can happen if a port is removed while
3810 * traffic is being received. Print a rate-limited message
3811 * in case it happens frequently. Install a drop flow so
3812 * that future packets of the flow are inexpensively dropped
3814 VLOG_INFO_RL(&rl, "received packet on unassociated port %"PRIu32,
3817 drop_key = drop_key_lookup(backer, upcall->key, upcall->key_len);
3819 drop_key = xmalloc(sizeof *drop_key);
3820 drop_key->key = xmemdup(upcall->key, upcall->key_len);
3821 drop_key->key_len = upcall->key_len;
3823 hmap_insert(&backer->drop_keys, &drop_key->hmap_node,
3824 hash_bytes(drop_key->key, drop_key->key_len, 0));
3825 dpif_flow_put(backer->dpif, DPIF_FP_CREATE | DPIF_FP_MODIFY,
3826 drop_key->key, drop_key->key_len, NULL, 0, NULL);
3834 ofproto->n_missed++;
3835 flow_extract(upcall->packet, flow.skb_priority, flow.skb_mark,
3836 &flow.tunnel, flow.in_port, &miss->flow);
3838 /* Add other packets to a to-do list. */
3839 hash = flow_hash(&miss->flow, 0);
3840 existing_miss = flow_miss_find(&todo, ofproto, &miss->flow, hash);
3841 if (!existing_miss) {
3842 hmap_insert(&todo, &miss->hmap_node, hash);
3843 miss->ofproto = ofproto;
3844 miss->key = upcall->key;
3845 miss->key_len = upcall->key_len;
3846 miss->upcall_type = upcall->type;
3847 miss->odp_in_port = odp_in_port;
3848 list_init(&miss->packets);
3852 miss = existing_miss;
3854 list_push_back(&miss->packets, &upcall->packet->list_node);
3857 /* Process each element in the to-do list, constructing the set of
3858 * operations to batch. */
3860 HMAP_FOR_EACH (miss, hmap_node, &todo) {
3861 handle_flow_miss(miss, flow_miss_ops, &n_ops);
3863 ovs_assert(n_ops <= ARRAY_SIZE(flow_miss_ops));
3865 /* Execute batch. */
3866 for (i = 0; i < n_ops; i++) {
3867 dpif_ops[i] = &flow_miss_ops[i].dpif_op;
3869 dpif_operate(backer->dpif, dpif_ops, n_ops);
3872 for (i = 0; i < n_ops; i++) {
3873 free(flow_miss_ops[i].garbage);
3875 hmap_destroy(&todo);
3878 static enum { SFLOW_UPCALL, MISS_UPCALL, BAD_UPCALL }
3879 classify_upcall(const struct dpif_upcall *upcall)
3881 union user_action_cookie cookie;
3883 /* First look at the upcall type. */
3884 switch (upcall->type) {
3885 case DPIF_UC_ACTION:
3891 case DPIF_N_UC_TYPES:
3893 VLOG_WARN_RL(&rl, "upcall has unexpected type %"PRIu32, upcall->type);
3897 /* "action" upcalls need a closer look. */
3898 if (!upcall->userdata) {
3899 VLOG_WARN_RL(&rl, "action upcall missing cookie");
3902 if (nl_attr_get_size(upcall->userdata) != sizeof(cookie)) {
3903 VLOG_WARN_RL(&rl, "action upcall cookie has unexpected size %zu",
3904 nl_attr_get_size(upcall->userdata));
3907 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof(cookie));
3908 switch (cookie.type) {
3909 case USER_ACTION_COOKIE_SFLOW:
3910 return SFLOW_UPCALL;
3912 case USER_ACTION_COOKIE_SLOW_PATH:
3915 case USER_ACTION_COOKIE_UNSPEC:
3917 VLOG_WARN_RL(&rl, "invalid user cookie : 0x%"PRIx64,
3918 nl_attr_get_u64(upcall->userdata));
3924 handle_sflow_upcall(struct dpif_backer *backer,
3925 const struct dpif_upcall *upcall)
3927 struct ofproto_dpif *ofproto;
3928 union user_action_cookie cookie;
3930 uint32_t odp_in_port;
3932 if (ofproto_receive(backer, upcall->packet, upcall->key, upcall->key_len,
3933 &flow, NULL, &ofproto, &odp_in_port, NULL)
3934 || !ofproto->sflow) {
3938 memcpy(&cookie, nl_attr_get(upcall->userdata), sizeof(cookie));
3939 dpif_sflow_received(ofproto->sflow, upcall->packet, &flow,
3940 odp_in_port, &cookie);
3944 handle_upcalls(struct dpif_backer *backer, unsigned int max_batch)
3946 struct dpif_upcall misses[FLOW_MISS_MAX_BATCH];
3947 struct ofpbuf miss_bufs[FLOW_MISS_MAX_BATCH];
3948 uint64_t miss_buf_stubs[FLOW_MISS_MAX_BATCH][4096 / 8];
3953 ovs_assert(max_batch <= FLOW_MISS_MAX_BATCH);
3956 for (n_processed = 0; n_processed < max_batch; n_processed++) {
3957 struct dpif_upcall *upcall = &misses[n_misses];
3958 struct ofpbuf *buf = &miss_bufs[n_misses];
3961 ofpbuf_use_stub(buf, miss_buf_stubs[n_misses],
3962 sizeof miss_buf_stubs[n_misses]);
3963 error = dpif_recv(backer->dpif, upcall, buf);
3969 switch (classify_upcall(upcall)) {
3971 /* Handle it later. */
3976 handle_sflow_upcall(backer, upcall);
3986 /* Handle deferred MISS_UPCALL processing. */
3987 handle_miss_upcalls(backer, misses, n_misses);
3988 for (i = 0; i < n_misses; i++) {
3989 ofpbuf_uninit(&miss_bufs[i]);
3995 /* Flow expiration. */
3997 static int subfacet_max_idle(const struct ofproto_dpif *);
3998 static void update_stats(struct dpif_backer *);
3999 static void rule_expire(struct rule_dpif *);
4000 static void expire_subfacets(struct ofproto_dpif *, int dp_max_idle);
4002 /* This function is called periodically by run(). Its job is to collect
4003 * updates for the flows that have been installed into the datapath, most
4004 * importantly when they last were used, and then use that information to
4005 * expire flows that have not been used recently.
4007 * Returns the number of milliseconds after which it should be called again. */
4009 expire(struct dpif_backer *backer)
4011 struct ofproto_dpif *ofproto;
4012 int max_idle = INT32_MAX;
4014 /* Periodically clear out the drop keys in an effort to keep them
4015 * relatively few. */
4016 drop_key_clear(backer);
4018 /* Update stats for each flow in the backer. */
4019 update_stats(backer);
4021 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
4022 struct rule *rule, *next_rule;
4025 if (ofproto->backer != backer) {
4029 /* Expire subfacets that have been idle too long. */
4030 dp_max_idle = subfacet_max_idle(ofproto);
4031 expire_subfacets(ofproto, dp_max_idle);
4033 max_idle = MIN(max_idle, dp_max_idle);
4035 /* Expire OpenFlow flows whose idle_timeout or hard_timeout
4037 LIST_FOR_EACH_SAFE (rule, next_rule, expirable,
4038 &ofproto->up.expirable) {
4039 rule_expire(rule_dpif_cast(rule));
4042 /* All outstanding data in existing flows has been accounted, so it's a
4043 * good time to do bond rebalancing. */
4044 if (ofproto->has_bonded_bundles) {
4045 struct ofbundle *bundle;
4047 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
4049 bond_rebalance(bundle->bond, &backer->revalidate_set);
4055 return MIN(max_idle, 1000);
4058 /* Updates flow table statistics given that the datapath just reported 'stats'
4059 * as 'subfacet''s statistics. */
4061 update_subfacet_stats(struct subfacet *subfacet,
4062 const struct dpif_flow_stats *stats)
4064 struct facet *facet = subfacet->facet;
4066 if (stats->n_packets >= subfacet->dp_packet_count) {
4067 uint64_t extra = stats->n_packets - subfacet->dp_packet_count;
4068 facet->packet_count += extra;
4070 VLOG_WARN_RL(&rl, "unexpected packet count from the datapath");
4073 if (stats->n_bytes >= subfacet->dp_byte_count) {
4074 facet->byte_count += stats->n_bytes - subfacet->dp_byte_count;
4076 VLOG_WARN_RL(&rl, "unexpected byte count from datapath");
4079 subfacet->dp_packet_count = stats->n_packets;
4080 subfacet->dp_byte_count = stats->n_bytes;
4082 facet->tcp_flags |= stats->tcp_flags;
4084 subfacet_update_time(subfacet, stats->used);
4085 if (facet->accounted_bytes < facet->byte_count) {
4087 facet_account(facet);
4088 facet->accounted_bytes = facet->byte_count;
4090 facet_push_stats(facet);
4093 /* 'key' with length 'key_len' bytes is a flow in 'dpif' that we know nothing
4094 * about, or a flow that shouldn't be installed but was anyway. Delete it. */
4096 delete_unexpected_flow(struct ofproto_dpif *ofproto,
4097 const struct nlattr *key, size_t key_len)
4099 if (!VLOG_DROP_WARN(&rl)) {
4103 odp_flow_key_format(key, key_len, &s);
4104 VLOG_WARN("unexpected flow on %s: %s", ofproto->up.name, ds_cstr(&s));
4108 COVERAGE_INC(facet_unexpected);
4109 dpif_flow_del(ofproto->backer->dpif, key, key_len, NULL);
4112 /* Update 'packet_count', 'byte_count', and 'used' members of installed facets.
4114 * This function also pushes statistics updates to rules which each facet
4115 * resubmits into. Generally these statistics will be accurate. However, if a
4116 * facet changes the rule it resubmits into at some time in between
4117 * update_stats() runs, it is possible that statistics accrued to the
4118 * old rule will be incorrectly attributed to the new rule. This could be
4119 * avoided by calling update_stats() whenever rules are created or
4120 * deleted. However, the performance impact of making so many calls to the
4121 * datapath do not justify the benefit of having perfectly accurate statistics.
4123 * In addition, this function maintains per ofproto flow hit counts. The patch
4124 * port is not treated specially. e.g. A packet ingress from br0 patched into
4125 * br1 will increase the hit count of br0 by 1, however, does not affect
4126 * the hit or miss counts of br1.
4129 update_stats(struct dpif_backer *backer)
4131 const struct dpif_flow_stats *stats;
4132 struct dpif_flow_dump dump;
4133 const struct nlattr *key;
4136 dpif_flow_dump_start(&dump, backer->dpif);
4137 while (dpif_flow_dump_next(&dump, &key, &key_len, NULL, NULL, &stats)) {
4139 struct subfacet *subfacet;
4140 struct ofproto_dpif *ofproto;
4141 struct ofport_dpif *ofport;
4144 if (ofproto_receive(backer, NULL, key, key_len, &flow, NULL, &ofproto,
4149 ofport = get_ofp_port(ofproto, flow.in_port);
4150 if (ofport && ofport->tnl_port) {
4151 netdev_vport_inc_rx(ofport->up.netdev, stats);
4154 key_hash = odp_flow_key_hash(key, key_len);
4155 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
4156 switch (subfacet ? subfacet->path : SF_NOT_INSTALLED) {
4158 /* Update ofproto_dpif's hit count. */
4159 if (stats->n_packets > subfacet->dp_packet_count) {
4160 uint64_t delta = stats->n_packets - subfacet->dp_packet_count;
4161 dpif_stats_update_hit_count(ofproto, delta);
4164 update_subfacet_stats(subfacet, stats);
4168 /* Stats are updated per-packet. */
4171 case SF_NOT_INSTALLED:
4173 delete_unexpected_flow(ofproto, key, key_len);
4177 dpif_flow_dump_done(&dump);
4180 /* Calculates and returns the number of milliseconds of idle time after which
4181 * subfacets should expire from the datapath. When a subfacet expires, we fold
4182 * its statistics into its facet, and when a facet's last subfacet expires, we
4183 * fold its statistic into its rule. */
4185 subfacet_max_idle(const struct ofproto_dpif *ofproto)
4188 * Idle time histogram.
4190 * Most of the time a switch has a relatively small number of subfacets.
4191 * When this is the case we might as well keep statistics for all of them
4192 * in userspace and to cache them in the kernel datapath for performance as
4195 * As the number of subfacets increases, the memory required to maintain
4196 * statistics about them in userspace and in the kernel becomes
4197 * significant. However, with a large number of subfacets it is likely
4198 * that only a few of them are "heavy hitters" that consume a large amount
4199 * of bandwidth. At this point, only heavy hitters are worth caching in
4200 * the kernel and maintaining in userspaces; other subfacets we can
4203 * The technique used to compute the idle time is to build a histogram with
4204 * N_BUCKETS buckets whose width is BUCKET_WIDTH msecs each. Each subfacet
4205 * that is installed in the kernel gets dropped in the appropriate bucket.
4206 * After the histogram has been built, we compute the cutoff so that only
4207 * the most-recently-used 1% of subfacets (but at least
4208 * ofproto->up.flow_eviction_threshold flows) are kept cached. At least
4209 * the most-recently-used bucket of subfacets is kept, so actually an
4210 * arbitrary number of subfacets can be kept in any given expiration run
4211 * (though the next run will delete most of those unless they receive
4214 * This requires a second pass through the subfacets, in addition to the
4215 * pass made by update_stats(), because the former function never looks at
4216 * uninstallable subfacets.
4218 enum { BUCKET_WIDTH = ROUND_UP(100, TIME_UPDATE_INTERVAL) };
4219 enum { N_BUCKETS = 5000 / BUCKET_WIDTH };
4220 int buckets[N_BUCKETS] = { 0 };
4221 int total, subtotal, bucket;
4222 struct subfacet *subfacet;
4226 total = hmap_count(&ofproto->subfacets);
4227 if (total <= ofproto->up.flow_eviction_threshold) {
4228 return N_BUCKETS * BUCKET_WIDTH;
4231 /* Build histogram. */
4233 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
4234 long long int idle = now - subfacet->used;
4235 int bucket = (idle <= 0 ? 0
4236 : idle >= BUCKET_WIDTH * N_BUCKETS ? N_BUCKETS - 1
4237 : (unsigned int) idle / BUCKET_WIDTH);
4241 /* Find the first bucket whose flows should be expired. */
4242 subtotal = bucket = 0;
4244 subtotal += buckets[bucket++];
4245 } while (bucket < N_BUCKETS &&
4246 subtotal < MAX(ofproto->up.flow_eviction_threshold, total / 100));
4248 if (VLOG_IS_DBG_ENABLED()) {
4252 ds_put_cstr(&s, "keep");
4253 for (i = 0; i < N_BUCKETS; i++) {
4255 ds_put_cstr(&s, ", drop");
4258 ds_put_format(&s, " %d:%d", i * BUCKET_WIDTH, buckets[i]);
4261 VLOG_INFO("%s: %s (msec:count)", ofproto->up.name, ds_cstr(&s));
4265 return bucket * BUCKET_WIDTH;
4269 expire_subfacets(struct ofproto_dpif *ofproto, int dp_max_idle)
4271 /* Cutoff time for most flows. */
4272 long long int normal_cutoff = time_msec() - dp_max_idle;
4274 /* We really want to keep flows for special protocols around, so use a more
4275 * conservative cutoff. */
4276 long long int special_cutoff = time_msec() - 10000;
4278 struct subfacet *subfacet, *next_subfacet;
4279 struct subfacet *batch[SUBFACET_DESTROY_MAX_BATCH];
4283 HMAP_FOR_EACH_SAFE (subfacet, next_subfacet, hmap_node,
4284 &ofproto->subfacets) {
4285 long long int cutoff;
4287 cutoff = (subfacet->slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)
4290 if (subfacet->used < cutoff) {
4291 if (subfacet->path != SF_NOT_INSTALLED) {
4292 batch[n_batch++] = subfacet;
4293 if (n_batch >= SUBFACET_DESTROY_MAX_BATCH) {
4294 subfacet_destroy_batch(ofproto, batch, n_batch);
4298 subfacet_destroy(subfacet);
4304 subfacet_destroy_batch(ofproto, batch, n_batch);
4308 /* If 'rule' is an OpenFlow rule, that has expired according to OpenFlow rules,
4309 * then delete it entirely. */
4311 rule_expire(struct rule_dpif *rule)
4313 struct facet *facet, *next_facet;
4317 if (rule->up.pending) {
4318 /* We'll have to expire it later. */
4322 /* Has 'rule' expired? */
4324 if (rule->up.hard_timeout
4325 && now > rule->up.modified + rule->up.hard_timeout * 1000) {
4326 reason = OFPRR_HARD_TIMEOUT;
4327 } else if (rule->up.idle_timeout
4328 && now > rule->up.used + rule->up.idle_timeout * 1000) {
4329 reason = OFPRR_IDLE_TIMEOUT;
4334 COVERAGE_INC(ofproto_dpif_expired);
4336 /* Update stats. (This is a no-op if the rule expired due to an idle
4337 * timeout, because that only happens when the rule has no facets left.) */
4338 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
4339 facet_remove(facet);
4342 /* Get rid of the rule. */
4343 ofproto_rule_expire(&rule->up, reason);
4348 /* Creates and returns a new facet owned by 'rule', given a 'flow'.
4350 * The caller must already have determined that no facet with an identical
4351 * 'flow' exists in 'ofproto' and that 'flow' is the best match for 'rule' in
4352 * the ofproto's classifier table.
4354 * 'hash' must be the return value of flow_hash(flow, 0).
4356 * The facet will initially have no subfacets. The caller should create (at
4357 * least) one subfacet with subfacet_create(). */
4358 static struct facet *
4359 facet_create(struct rule_dpif *rule, const struct flow *flow, uint32_t hash)
4361 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4362 struct facet *facet;
4364 facet = xzalloc(sizeof *facet);
4365 facet->used = time_msec();
4366 hmap_insert(&ofproto->facets, &facet->hmap_node, hash);
4367 list_push_back(&rule->facets, &facet->list_node);
4369 facet->flow = *flow;
4370 list_init(&facet->subfacets);
4371 netflow_flow_init(&facet->nf_flow);
4372 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, facet->used);
4378 facet_free(struct facet *facet)
4383 /* Executes, within 'ofproto', the 'n_actions' actions in 'actions' on
4384 * 'packet', which arrived on 'in_port'. */
4386 execute_odp_actions(struct ofproto_dpif *ofproto, const struct flow *flow,
4387 const struct nlattr *odp_actions, size_t actions_len,
4388 struct ofpbuf *packet)
4390 struct odputil_keybuf keybuf;
4394 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
4395 odp_flow_key_from_flow(&key, flow,
4396 ofp_port_to_odp_port(ofproto, flow->in_port));
4398 error = dpif_execute(ofproto->backer->dpif, key.data, key.size,
4399 odp_actions, actions_len, packet);
4403 /* Remove 'facet' from 'ofproto' and free up the associated memory:
4405 * - If 'facet' was installed in the datapath, uninstalls it and updates its
4406 * rule's statistics, via subfacet_uninstall().
4408 * - Removes 'facet' from its rule and from ofproto->facets.
4411 facet_remove(struct facet *facet)
4413 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4414 struct subfacet *subfacet, *next_subfacet;
4416 ovs_assert(!list_is_empty(&facet->subfacets));
4418 /* First uninstall all of the subfacets to get final statistics. */
4419 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4420 subfacet_uninstall(subfacet);
4423 /* Flush the final stats to the rule.
4425 * This might require us to have at least one subfacet around so that we
4426 * can use its actions for accounting in facet_account(), which is why we
4427 * have uninstalled but not yet destroyed the subfacets. */
4428 facet_flush_stats(facet);
4430 /* Now we're really all done so destroy everything. */
4431 LIST_FOR_EACH_SAFE (subfacet, next_subfacet, list_node,
4432 &facet->subfacets) {
4433 subfacet_destroy__(subfacet);
4435 hmap_remove(&ofproto->facets, &facet->hmap_node);
4436 list_remove(&facet->list_node);
4440 /* Feed information from 'facet' back into the learning table to keep it in
4441 * sync with what is actually flowing through the datapath. */
4443 facet_learn(struct facet *facet)
4445 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4446 struct subfacet *subfacet= CONTAINER_OF(list_front(&facet->subfacets),
4447 struct subfacet, list_node);
4448 struct action_xlate_ctx ctx;
4450 if (!facet->has_learn
4451 && !facet->has_normal
4452 && (!facet->has_fin_timeout
4453 || !(facet->tcp_flags & (TCP_FIN | TCP_RST)))) {
4457 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4458 &subfacet->initial_vals,
4459 facet->rule, facet->tcp_flags, NULL);
4460 ctx.may_learn = true;
4461 xlate_actions_for_side_effects(&ctx, facet->rule->up.ofpacts,
4462 facet->rule->up.ofpacts_len);
4466 facet_account(struct facet *facet)
4468 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4469 struct subfacet *subfacet = facet_get_subfacet(facet);
4470 const struct nlattr *a;
4475 if (!facet->has_normal || !ofproto->has_bonded_bundles) {
4478 n_bytes = facet->byte_count - facet->accounted_bytes;
4480 /* This loop feeds byte counters to bond_account() for rebalancing to use
4481 * as a basis. We also need to track the actual VLAN on which the packet
4482 * is going to be sent to ensure that it matches the one passed to
4483 * bond_choose_output_slave(). (Otherwise, we will account to the wrong
4486 * We use the actions from an arbitrary subfacet because they should all
4487 * be equally valid for our purpose. */
4488 vlan_tci = facet->flow.vlan_tci;
4489 NL_ATTR_FOR_EACH_UNSAFE (a, left,
4490 subfacet->actions, subfacet->actions_len) {
4491 const struct ovs_action_push_vlan *vlan;
4492 struct ofport_dpif *port;
4494 switch (nl_attr_type(a)) {
4495 case OVS_ACTION_ATTR_OUTPUT:
4496 port = get_odp_port(ofproto, nl_attr_get_u32(a));
4497 if (port && port->bundle && port->bundle->bond) {
4498 bond_account(port->bundle->bond, &facet->flow,
4499 vlan_tci_to_vid(vlan_tci), n_bytes);
4503 case OVS_ACTION_ATTR_POP_VLAN:
4504 vlan_tci = htons(0);
4507 case OVS_ACTION_ATTR_PUSH_VLAN:
4508 vlan = nl_attr_get(a);
4509 vlan_tci = vlan->vlan_tci;
4515 /* Returns true if the only action for 'facet' is to send to the controller.
4516 * (We don't report NetFlow expiration messages for such facets because they
4517 * are just part of the control logic for the network, not real traffic). */
4519 facet_is_controller_flow(struct facet *facet)
4522 const struct rule *rule = &facet->rule->up;
4523 const struct ofpact *ofpacts = rule->ofpacts;
4524 size_t ofpacts_len = rule->ofpacts_len;
4526 if (ofpacts_len > 0 &&
4527 ofpacts->type == OFPACT_CONTROLLER &&
4528 ofpact_next(ofpacts) >= ofpact_end(ofpacts, ofpacts_len)) {
4535 /* Folds all of 'facet''s statistics into its rule. Also updates the
4536 * accounting ofhook and emits a NetFlow expiration if appropriate. All of
4537 * 'facet''s statistics in the datapath should have been zeroed and folded into
4538 * its packet and byte counts before this function is called. */
4540 facet_flush_stats(struct facet *facet)
4542 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4543 struct subfacet *subfacet;
4545 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4546 ovs_assert(!subfacet->dp_byte_count);
4547 ovs_assert(!subfacet->dp_packet_count);
4550 facet_push_stats(facet);
4551 if (facet->accounted_bytes < facet->byte_count) {
4552 facet_account(facet);
4553 facet->accounted_bytes = facet->byte_count;
4556 if (ofproto->netflow && !facet_is_controller_flow(facet)) {
4557 struct ofexpired expired;
4558 expired.flow = facet->flow;
4559 expired.packet_count = facet->packet_count;
4560 expired.byte_count = facet->byte_count;
4561 expired.used = facet->used;
4562 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
4565 facet->rule->packet_count += facet->packet_count;
4566 facet->rule->byte_count += facet->byte_count;
4568 /* Reset counters to prevent double counting if 'facet' ever gets
4570 facet_reset_counters(facet);
4572 netflow_flow_clear(&facet->nf_flow);
4573 facet->tcp_flags = 0;
4576 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4577 * Returns it if found, otherwise a null pointer.
4579 * 'hash' must be the return value of flow_hash(flow, 0).
4581 * The returned facet might need revalidation; use facet_lookup_valid()
4582 * instead if that is important. */
4583 static struct facet *
4584 facet_find(struct ofproto_dpif *ofproto,
4585 const struct flow *flow, uint32_t hash)
4587 struct facet *facet;
4589 HMAP_FOR_EACH_WITH_HASH (facet, hmap_node, hash, &ofproto->facets) {
4590 if (flow_equal(flow, &facet->flow)) {
4598 /* Searches 'ofproto''s table of facets for one exactly equal to 'flow'.
4599 * Returns it if found, otherwise a null pointer.
4601 * 'hash' must be the return value of flow_hash(flow, 0).
4603 * The returned facet is guaranteed to be valid. */
4604 static struct facet *
4605 facet_lookup_valid(struct ofproto_dpif *ofproto, const struct flow *flow,
4608 struct facet *facet;
4610 facet = facet_find(ofproto, flow, hash);
4612 && (ofproto->backer->need_revalidate
4613 || tag_set_intersects(&ofproto->backer->revalidate_set,
4615 facet_revalidate(facet);
4617 /* facet_revalidate() may have destroyed 'facet'. */
4618 facet = facet_find(ofproto, flow, hash);
4624 /* Return a subfacet from 'facet'. A facet consists of one or more
4625 * subfacets, and this function returns one of them. */
4626 static struct subfacet *facet_get_subfacet(struct facet *facet)
4628 return CONTAINER_OF(list_front(&facet->subfacets), struct subfacet,
4633 subfacet_path_to_string(enum subfacet_path path)
4636 case SF_NOT_INSTALLED:
4637 return "not installed";
4639 return "in fast path";
4641 return "in slow path";
4647 /* Returns the path in which a subfacet should be installed if its 'slow'
4648 * member has the specified value. */
4649 static enum subfacet_path
4650 subfacet_want_path(enum slow_path_reason slow)
4652 return slow ? SF_SLOW_PATH : SF_FAST_PATH;
4655 /* Returns true if 'subfacet' needs to have its datapath flow updated,
4656 * supposing that its actions have been recalculated as 'want_actions' and that
4657 * 'slow' is nonzero iff 'subfacet' should be in the slow path. */
4659 subfacet_should_install(struct subfacet *subfacet, enum slow_path_reason slow,
4660 const struct ofpbuf *want_actions)
4662 enum subfacet_path want_path = subfacet_want_path(slow);
4663 return (want_path != subfacet->path
4664 || (want_path == SF_FAST_PATH
4665 && (subfacet->actions_len != want_actions->size
4666 || memcmp(subfacet->actions, want_actions->data,
4667 subfacet->actions_len))));
4671 facet_check_consistency(struct facet *facet)
4673 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 15);
4675 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4677 uint64_t odp_actions_stub[1024 / 8];
4678 struct ofpbuf odp_actions;
4680 struct rule_dpif *rule;
4681 struct subfacet *subfacet;
4682 bool may_log = false;
4685 /* Check the rule for consistency. */
4686 rule = rule_dpif_lookup(ofproto, &facet->flow);
4687 ok = rule == facet->rule;
4689 may_log = !VLOG_DROP_WARN(&rl);
4694 flow_format(&s, &facet->flow);
4695 ds_put_format(&s, ": facet associated with wrong rule (was "
4696 "table=%"PRIu8",", facet->rule->up.table_id);
4697 cls_rule_format(&facet->rule->up.cr, &s);
4698 ds_put_format(&s, ") (should have been table=%"PRIu8",",
4700 cls_rule_format(&rule->up.cr, &s);
4701 ds_put_char(&s, ')');
4703 VLOG_WARN("%s", ds_cstr(&s));
4708 /* Check the datapath actions for consistency. */
4709 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4710 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4711 enum subfacet_path want_path;
4712 struct action_xlate_ctx ctx;
4715 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4716 &subfacet->initial_vals, rule, 0, NULL);
4717 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len,
4720 if (subfacet->path == SF_NOT_INSTALLED) {
4721 /* This only happens if the datapath reported an error when we
4722 * tried to install the flow. Don't flag another error here. */
4726 want_path = subfacet_want_path(subfacet->slow);
4727 if (want_path == SF_SLOW_PATH && subfacet->path == SF_SLOW_PATH) {
4728 /* The actions for slow-path flows may legitimately vary from one
4729 * packet to the next. We're done. */
4733 if (!subfacet_should_install(subfacet, subfacet->slow, &odp_actions)) {
4737 /* Inconsistency! */
4739 may_log = !VLOG_DROP_WARN(&rl);
4743 /* Rate-limited, skip reporting. */
4748 odp_flow_key_format(subfacet->key, subfacet->key_len, &s);
4750 ds_put_cstr(&s, ": inconsistency in subfacet");
4751 if (want_path != subfacet->path) {
4752 enum odp_key_fitness fitness = subfacet->key_fitness;
4754 ds_put_format(&s, " (%s, fitness=%s)",
4755 subfacet_path_to_string(subfacet->path),
4756 odp_key_fitness_to_string(fitness));
4757 ds_put_format(&s, " (should have been %s)",
4758 subfacet_path_to_string(want_path));
4759 } else if (want_path == SF_FAST_PATH) {
4760 ds_put_cstr(&s, " (actions were: ");
4761 format_odp_actions(&s, subfacet->actions,
4762 subfacet->actions_len);
4763 ds_put_cstr(&s, ") (correct actions: ");
4764 format_odp_actions(&s, odp_actions.data, odp_actions.size);
4765 ds_put_char(&s, ')');
4767 ds_put_cstr(&s, " (actions: ");
4768 format_odp_actions(&s, subfacet->actions,
4769 subfacet->actions_len);
4770 ds_put_char(&s, ')');
4772 VLOG_WARN("%s", ds_cstr(&s));
4775 ofpbuf_uninit(&odp_actions);
4780 /* Re-searches the classifier for 'facet':
4782 * - If the rule found is different from 'facet''s current rule, moves
4783 * 'facet' to the new rule and recompiles its actions.
4785 * - If the rule found is the same as 'facet''s current rule, leaves 'facet'
4786 * where it is and recompiles its actions anyway.
4788 * - If any of 'facet''s subfacets correspond to a new flow according to
4789 * ofproto_receive(), 'facet' is removed. */
4791 facet_revalidate(struct facet *facet)
4793 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4795 struct nlattr *odp_actions;
4798 struct actions *new_actions;
4800 struct action_xlate_ctx ctx;
4801 uint64_t odp_actions_stub[1024 / 8];
4802 struct ofpbuf odp_actions;
4804 struct rule_dpif *new_rule;
4805 struct subfacet *subfacet;
4808 COVERAGE_INC(facet_revalidate);
4810 /* Check that child subfacets still correspond to this facet. Tunnel
4811 * configuration changes could cause a subfacet's OpenFlow in_port to
4813 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4814 struct ofproto_dpif *recv_ofproto;
4815 struct flow recv_flow;
4818 error = ofproto_receive(ofproto->backer, NULL, subfacet->key,
4819 subfacet->key_len, &recv_flow, NULL,
4820 &recv_ofproto, NULL, NULL);
4822 || recv_ofproto != ofproto
4823 || memcmp(&recv_flow, &facet->flow, sizeof recv_flow)) {
4824 facet_remove(facet);
4829 new_rule = rule_dpif_lookup(ofproto, &facet->flow);
4831 /* Calculate new datapath actions.
4833 * We do not modify any 'facet' state yet, because we might need to, e.g.,
4834 * emit a NetFlow expiration and, if so, we need to have the old state
4835 * around to properly compose it. */
4837 /* If the datapath actions changed or the installability changed,
4838 * then we need to talk to the datapath. */
4841 memset(&ctx, 0, sizeof ctx);
4842 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
4843 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4844 enum slow_path_reason slow;
4846 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4847 &subfacet->initial_vals, new_rule, 0, NULL);
4848 xlate_actions(&ctx, new_rule->up.ofpacts, new_rule->up.ofpacts_len,
4851 slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4852 if (subfacet_should_install(subfacet, slow, &odp_actions)) {
4853 struct dpif_flow_stats stats;
4855 subfacet_install(subfacet,
4856 odp_actions.data, odp_actions.size, &stats, slow);
4857 subfacet_update_stats(subfacet, &stats);
4860 new_actions = xcalloc(list_size(&facet->subfacets),
4861 sizeof *new_actions);
4863 new_actions[i].odp_actions = xmemdup(odp_actions.data,
4865 new_actions[i].actions_len = odp_actions.size;
4870 ofpbuf_uninit(&odp_actions);
4873 facet_flush_stats(facet);
4876 /* Update 'facet' now that we've taken care of all the old state. */
4877 facet->tags = ctx.tags;
4878 facet->nf_flow.output_iface = ctx.nf_output_iface;
4879 facet->has_learn = ctx.has_learn;
4880 facet->has_normal = ctx.has_normal;
4881 facet->has_fin_timeout = ctx.has_fin_timeout;
4882 facet->mirrors = ctx.mirrors;
4885 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
4886 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
4888 if (new_actions && new_actions[i].odp_actions) {
4889 free(subfacet->actions);
4890 subfacet->actions = new_actions[i].odp_actions;
4891 subfacet->actions_len = new_actions[i].actions_len;
4897 if (facet->rule != new_rule) {
4898 COVERAGE_INC(facet_changed_rule);
4899 list_remove(&facet->list_node);
4900 list_push_back(&new_rule->facets, &facet->list_node);
4901 facet->rule = new_rule;
4902 facet->used = new_rule->up.created;
4903 facet->prev_used = facet->used;
4907 /* Updates 'facet''s used time. Caller is responsible for calling
4908 * facet_push_stats() to update the flows which 'facet' resubmits into. */
4910 facet_update_time(struct facet *facet, long long int used)
4912 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
4913 if (used > facet->used) {
4915 ofproto_rule_update_used(&facet->rule->up, used);
4916 netflow_flow_update_time(ofproto->netflow, &facet->nf_flow, used);
4921 facet_reset_counters(struct facet *facet)
4923 facet->packet_count = 0;
4924 facet->byte_count = 0;
4925 facet->prev_packet_count = 0;
4926 facet->prev_byte_count = 0;
4927 facet->accounted_bytes = 0;
4931 facet_push_stats(struct facet *facet)
4933 struct dpif_flow_stats stats;
4935 ovs_assert(facet->packet_count >= facet->prev_packet_count);
4936 ovs_assert(facet->byte_count >= facet->prev_byte_count);
4937 ovs_assert(facet->used >= facet->prev_used);
4939 stats.n_packets = facet->packet_count - facet->prev_packet_count;
4940 stats.n_bytes = facet->byte_count - facet->prev_byte_count;
4941 stats.used = facet->used;
4942 stats.tcp_flags = 0;
4944 if (stats.n_packets || stats.n_bytes || facet->used > facet->prev_used) {
4945 facet->prev_packet_count = facet->packet_count;
4946 facet->prev_byte_count = facet->byte_count;
4947 facet->prev_used = facet->used;
4949 flow_push_stats(facet, &stats);
4951 update_mirror_stats(ofproto_dpif_cast(facet->rule->up.ofproto),
4952 facet->mirrors, stats.n_packets, stats.n_bytes);
4957 rule_credit_stats(struct rule_dpif *rule, const struct dpif_flow_stats *stats)
4959 rule->packet_count += stats->n_packets;
4960 rule->byte_count += stats->n_bytes;
4961 ofproto_rule_update_used(&rule->up, stats->used);
4964 /* Pushes flow statistics to the rules which 'facet->flow' resubmits
4965 * into given 'facet->rule''s actions and mirrors. */
4967 flow_push_stats(struct facet *facet, const struct dpif_flow_stats *stats)
4969 struct rule_dpif *rule = facet->rule;
4970 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
4971 struct subfacet *subfacet = facet_get_subfacet(facet);
4972 struct action_xlate_ctx ctx;
4974 ofproto_rule_update_used(&rule->up, stats->used);
4976 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
4977 &subfacet->initial_vals, rule, 0, NULL);
4978 ctx.resubmit_stats = stats;
4979 xlate_actions_for_side_effects(&ctx, rule->up.ofpacts,
4980 rule->up.ofpacts_len);
4985 static struct subfacet *
4986 subfacet_find(struct ofproto_dpif *ofproto,
4987 const struct nlattr *key, size_t key_len, uint32_t key_hash)
4989 struct subfacet *subfacet;
4991 HMAP_FOR_EACH_WITH_HASH (subfacet, hmap_node, key_hash,
4992 &ofproto->subfacets) {
4993 if (subfacet->key_len == key_len
4994 && !memcmp(key, subfacet->key, key_len)) {
5002 /* Searches 'facet' (within 'ofproto') for a subfacet with the specified
5003 * 'key_fitness', 'key', and 'key_len' members in 'miss'. Returns the
5004 * existing subfacet if there is one, otherwise creates and returns a
5007 * If the returned subfacet is new, then subfacet->actions will be NULL, in
5008 * which case the caller must populate the actions with
5009 * subfacet_make_actions(). */
5010 static struct subfacet *
5011 subfacet_create(struct facet *facet, struct flow_miss *miss,
5014 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5015 enum odp_key_fitness key_fitness = miss->key_fitness;
5016 const struct nlattr *key = miss->key;
5017 size_t key_len = miss->key_len;
5019 struct subfacet *subfacet;
5021 key_hash = odp_flow_key_hash(key, key_len);
5023 if (list_is_empty(&facet->subfacets)) {
5024 subfacet = &facet->one_subfacet;
5026 subfacet = subfacet_find(ofproto, key, key_len, key_hash);
5028 if (subfacet->facet == facet) {
5032 /* This shouldn't happen. */
5033 VLOG_ERR_RL(&rl, "subfacet with wrong facet");
5034 subfacet_destroy(subfacet);
5037 subfacet = xmalloc(sizeof *subfacet);
5040 hmap_insert(&ofproto->subfacets, &subfacet->hmap_node, key_hash);
5041 list_push_back(&facet->subfacets, &subfacet->list_node);
5042 subfacet->facet = facet;
5043 subfacet->key_fitness = key_fitness;
5044 subfacet->key = xmemdup(key, key_len);
5045 subfacet->key_len = key_len;
5046 subfacet->used = now;
5047 subfacet->dp_packet_count = 0;
5048 subfacet->dp_byte_count = 0;
5049 subfacet->actions_len = 0;
5050 subfacet->actions = NULL;
5051 subfacet->slow = (subfacet->key_fitness == ODP_FIT_TOO_LITTLE
5054 subfacet->path = SF_NOT_INSTALLED;
5055 subfacet->initial_vals = miss->initial_vals;
5056 subfacet->odp_in_port = miss->odp_in_port;
5061 /* Uninstalls 'subfacet' from the datapath, if it is installed, removes it from
5062 * its facet within 'ofproto', and frees it. */
5064 subfacet_destroy__(struct subfacet *subfacet)
5066 struct facet *facet = subfacet->facet;
5067 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5069 subfacet_uninstall(subfacet);
5070 hmap_remove(&ofproto->subfacets, &subfacet->hmap_node);
5071 list_remove(&subfacet->list_node);
5072 free(subfacet->key);
5073 free(subfacet->actions);
5074 if (subfacet != &facet->one_subfacet) {
5079 /* Destroys 'subfacet', as with subfacet_destroy__(), and then if this was the
5080 * last remaining subfacet in its facet destroys the facet too. */
5082 subfacet_destroy(struct subfacet *subfacet)
5084 struct facet *facet = subfacet->facet;
5086 if (list_is_singleton(&facet->subfacets)) {
5087 /* facet_remove() needs at least one subfacet (it will remove it). */
5088 facet_remove(facet);
5090 subfacet_destroy__(subfacet);
5095 subfacet_destroy_batch(struct ofproto_dpif *ofproto,
5096 struct subfacet **subfacets, int n)
5098 struct dpif_op ops[SUBFACET_DESTROY_MAX_BATCH];
5099 struct dpif_op *opsp[SUBFACET_DESTROY_MAX_BATCH];
5100 struct dpif_flow_stats stats[SUBFACET_DESTROY_MAX_BATCH];
5103 for (i = 0; i < n; i++) {
5104 ops[i].type = DPIF_OP_FLOW_DEL;
5105 ops[i].u.flow_del.key = subfacets[i]->key;
5106 ops[i].u.flow_del.key_len = subfacets[i]->key_len;
5107 ops[i].u.flow_del.stats = &stats[i];
5111 dpif_operate(ofproto->backer->dpif, opsp, n);
5112 for (i = 0; i < n; i++) {
5113 subfacet_reset_dp_stats(subfacets[i], &stats[i]);
5114 subfacets[i]->path = SF_NOT_INSTALLED;
5115 subfacet_destroy(subfacets[i]);
5119 /* Composes the datapath actions for 'subfacet' based on its rule's actions.
5120 * Translates the actions into 'odp_actions', which the caller must have
5121 * initialized and is responsible for uninitializing. */
5123 subfacet_make_actions(struct subfacet *subfacet, const struct ofpbuf *packet,
5124 struct ofpbuf *odp_actions)
5126 struct facet *facet = subfacet->facet;
5127 struct rule_dpif *rule = facet->rule;
5128 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5130 struct action_xlate_ctx ctx;
5132 action_xlate_ctx_init(&ctx, ofproto, &facet->flow,
5133 &subfacet->initial_vals, rule, 0, packet);
5134 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, odp_actions);
5135 facet->tags = ctx.tags;
5136 facet->has_learn = ctx.has_learn;
5137 facet->has_normal = ctx.has_normal;
5138 facet->has_fin_timeout = ctx.has_fin_timeout;
5139 facet->nf_flow.output_iface = ctx.nf_output_iface;
5140 facet->mirrors = ctx.mirrors;
5142 subfacet->slow = (subfacet->slow & SLOW_MATCH) | ctx.slow;
5143 if (subfacet->actions_len != odp_actions->size
5144 || memcmp(subfacet->actions, odp_actions->data, odp_actions->size)) {
5145 free(subfacet->actions);
5146 subfacet->actions_len = odp_actions->size;
5147 subfacet->actions = xmemdup(odp_actions->data, odp_actions->size);
5151 /* Updates 'subfacet''s datapath flow, setting its actions to 'actions_len'
5152 * bytes of actions in 'actions'. If 'stats' is non-null, statistics counters
5153 * in the datapath will be zeroed and 'stats' will be updated with traffic new
5154 * since 'subfacet' was last updated.
5156 * Returns 0 if successful, otherwise a positive errno value. */
5158 subfacet_install(struct subfacet *subfacet,
5159 const struct nlattr *actions, size_t actions_len,
5160 struct dpif_flow_stats *stats,
5161 enum slow_path_reason slow)
5163 struct facet *facet = subfacet->facet;
5164 struct ofproto_dpif *ofproto = ofproto_dpif_cast(facet->rule->up.ofproto);
5165 enum subfacet_path path = subfacet_want_path(slow);
5166 uint64_t slow_path_stub[128 / 8];
5167 enum dpif_flow_put_flags flags;
5170 flags = DPIF_FP_CREATE | DPIF_FP_MODIFY;
5172 flags |= DPIF_FP_ZERO_STATS;
5175 if (path == SF_SLOW_PATH) {
5176 compose_slow_path(ofproto, &facet->flow, slow,
5177 slow_path_stub, sizeof slow_path_stub,
5178 &actions, &actions_len);
5181 ret = dpif_flow_put(ofproto->backer->dpif, flags, subfacet->key,
5182 subfacet->key_len, actions, actions_len, stats);
5185 subfacet_reset_dp_stats(subfacet, stats);
5189 subfacet->path = path;
5195 subfacet_reinstall(struct subfacet *subfacet, struct dpif_flow_stats *stats)
5197 return subfacet_install(subfacet, subfacet->actions, subfacet->actions_len,
5198 stats, subfacet->slow);
5201 /* If 'subfacet' is installed in the datapath, uninstalls it. */
5203 subfacet_uninstall(struct subfacet *subfacet)
5205 if (subfacet->path != SF_NOT_INSTALLED) {
5206 struct rule_dpif *rule = subfacet->facet->rule;
5207 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5208 struct dpif_flow_stats stats;
5211 error = dpif_flow_del(ofproto->backer->dpif, subfacet->key,
5212 subfacet->key_len, &stats);
5213 subfacet_reset_dp_stats(subfacet, &stats);
5215 subfacet_update_stats(subfacet, &stats);
5217 subfacet->path = SF_NOT_INSTALLED;
5219 ovs_assert(subfacet->dp_packet_count == 0);
5220 ovs_assert(subfacet->dp_byte_count == 0);
5224 /* Resets 'subfacet''s datapath statistics counters. This should be called
5225 * when 'subfacet''s statistics are cleared in the datapath. If 'stats' is
5226 * non-null, it should contain the statistics returned by dpif when 'subfacet'
5227 * was reset in the datapath. 'stats' will be modified to include only
5228 * statistics new since 'subfacet' was last updated. */
5230 subfacet_reset_dp_stats(struct subfacet *subfacet,
5231 struct dpif_flow_stats *stats)
5234 && subfacet->dp_packet_count <= stats->n_packets
5235 && subfacet->dp_byte_count <= stats->n_bytes) {
5236 stats->n_packets -= subfacet->dp_packet_count;
5237 stats->n_bytes -= subfacet->dp_byte_count;
5240 subfacet->dp_packet_count = 0;
5241 subfacet->dp_byte_count = 0;
5244 /* Updates 'subfacet''s used time. The caller is responsible for calling
5245 * facet_push_stats() to update the flows which 'subfacet' resubmits into. */
5247 subfacet_update_time(struct subfacet *subfacet, long long int used)
5249 if (used > subfacet->used) {
5250 subfacet->used = used;
5251 facet_update_time(subfacet->facet, used);
5255 /* Folds the statistics from 'stats' into the counters in 'subfacet'.
5257 * Because of the meaning of a subfacet's counters, it only makes sense to do
5258 * this if 'stats' are not tracked in the datapath, that is, if 'stats'
5259 * represents a packet that was sent by hand or if it represents statistics
5260 * that have been cleared out of the datapath. */
5262 subfacet_update_stats(struct subfacet *subfacet,
5263 const struct dpif_flow_stats *stats)
5265 if (stats->n_packets || stats->used > subfacet->used) {
5266 struct facet *facet = subfacet->facet;
5268 subfacet_update_time(subfacet, stats->used);
5269 facet->packet_count += stats->n_packets;
5270 facet->byte_count += stats->n_bytes;
5271 facet->tcp_flags |= stats->tcp_flags;
5272 facet_push_stats(facet);
5273 netflow_flow_update_flags(&facet->nf_flow, stats->tcp_flags);
5279 static struct rule_dpif *
5280 rule_dpif_lookup(struct ofproto_dpif *ofproto, const struct flow *flow)
5282 struct rule_dpif *rule;
5284 rule = rule_dpif_lookup__(ofproto, flow, 0);
5289 return rule_dpif_miss_rule(ofproto, flow);
5292 static struct rule_dpif *
5293 rule_dpif_lookup__(struct ofproto_dpif *ofproto, const struct flow *flow,
5296 struct cls_rule *cls_rule;
5297 struct classifier *cls;
5299 if (table_id >= N_TABLES) {
5303 cls = &ofproto->up.tables[table_id].cls;
5304 if (flow->nw_frag & FLOW_NW_FRAG_ANY
5305 && ofproto->up.frag_handling == OFPC_FRAG_NORMAL) {
5306 /* For OFPC_NORMAL frag_handling, we must pretend that transport ports
5307 * are unavailable. */
5308 struct flow ofpc_normal_flow = *flow;
5309 ofpc_normal_flow.tp_src = htons(0);
5310 ofpc_normal_flow.tp_dst = htons(0);
5311 cls_rule = classifier_lookup(cls, &ofpc_normal_flow);
5313 cls_rule = classifier_lookup(cls, flow);
5315 return rule_dpif_cast(rule_from_cls_rule(cls_rule));
5318 static struct rule_dpif *
5319 rule_dpif_miss_rule(struct ofproto_dpif *ofproto, const struct flow *flow)
5321 struct ofport_dpif *port;
5323 port = get_ofp_port(ofproto, flow->in_port);
5325 VLOG_WARN_RL(&rl, "packet-in on unknown port %"PRIu16, flow->in_port);
5326 return ofproto->miss_rule;
5329 if (port->up.pp.config & OFPUTIL_PC_NO_PACKET_IN) {
5330 return ofproto->no_packet_in_rule;
5332 return ofproto->miss_rule;
5336 complete_operation(struct rule_dpif *rule)
5338 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5340 rule_invalidate(rule);
5342 struct dpif_completion *c = xmalloc(sizeof *c);
5343 c->op = rule->up.pending;
5344 list_push_back(&ofproto->completions, &c->list_node);
5346 ofoperation_complete(rule->up.pending, 0);
5350 static struct rule *
5353 struct rule_dpif *rule = xmalloc(sizeof *rule);
5358 rule_dealloc(struct rule *rule_)
5360 struct rule_dpif *rule = rule_dpif_cast(rule_);
5365 rule_construct(struct rule *rule_)
5367 struct rule_dpif *rule = rule_dpif_cast(rule_);
5368 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5369 struct rule_dpif *victim;
5372 rule->packet_count = 0;
5373 rule->byte_count = 0;
5375 victim = rule_dpif_cast(ofoperation_get_victim(rule->up.pending));
5376 if (victim && !list_is_empty(&victim->facets)) {
5377 struct facet *facet;
5379 rule->facets = victim->facets;
5380 list_moved(&rule->facets);
5381 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5382 /* XXX: We're only clearing our local counters here. It's possible
5383 * that quite a few packets are unaccounted for in the datapath
5384 * statistics. These will be accounted to the new rule instead of
5385 * cleared as required. This could be fixed by clearing out the
5386 * datapath statistics for this facet, but currently it doesn't
5388 facet_reset_counters(facet);
5392 /* Must avoid list_moved() in this case. */
5393 list_init(&rule->facets);
5396 table_id = rule->up.table_id;
5398 rule->tag = victim->tag;
5399 } else if (table_id == 0) {
5404 miniflow_expand(&rule->up.cr.match.flow, &flow);
5405 rule->tag = rule_calculate_tag(&flow, &rule->up.cr.match.mask,
5406 ofproto->tables[table_id].basis);
5409 complete_operation(rule);
5414 rule_destruct(struct rule *rule_)
5416 struct rule_dpif *rule = rule_dpif_cast(rule_);
5417 struct facet *facet, *next_facet;
5419 LIST_FOR_EACH_SAFE (facet, next_facet, list_node, &rule->facets) {
5420 facet_revalidate(facet);
5423 complete_operation(rule);
5427 rule_get_stats(struct rule *rule_, uint64_t *packets, uint64_t *bytes)
5429 struct rule_dpif *rule = rule_dpif_cast(rule_);
5430 struct facet *facet;
5432 /* Start from historical data for 'rule' itself that are no longer tracked
5433 * in facets. This counts, for example, facets that have expired. */
5434 *packets = rule->packet_count;
5435 *bytes = rule->byte_count;
5437 /* Add any statistics that are tracked by facets. This includes
5438 * statistical data recently updated by ofproto_update_stats() as well as
5439 * stats for packets that were executed "by hand" via dpif_execute(). */
5440 LIST_FOR_EACH (facet, list_node, &rule->facets) {
5441 *packets += facet->packet_count;
5442 *bytes += facet->byte_count;
5447 rule_dpif_execute(struct rule_dpif *rule, const struct flow *flow,
5448 struct ofpbuf *packet)
5450 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
5451 struct initial_vals initial_vals;
5452 struct dpif_flow_stats stats;
5453 struct action_xlate_ctx ctx;
5454 uint64_t odp_actions_stub[1024 / 8];
5455 struct ofpbuf odp_actions;
5457 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
5458 rule_credit_stats(rule, &stats);
5460 initial_vals.vlan_tci = flow->vlan_tci;
5461 initial_vals.tunnel_ip_tos = flow->tunnel.ip_tos;
5462 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5463 action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals,
5464 rule, stats.tcp_flags, packet);
5465 ctx.resubmit_stats = &stats;
5466 xlate_actions(&ctx, rule->up.ofpacts, rule->up.ofpacts_len, &odp_actions);
5468 execute_odp_actions(ofproto, flow, odp_actions.data,
5469 odp_actions.size, packet);
5471 ofpbuf_uninit(&odp_actions);
5475 rule_execute(struct rule *rule, const struct flow *flow,
5476 struct ofpbuf *packet)
5478 rule_dpif_execute(rule_dpif_cast(rule), flow, packet);
5479 ofpbuf_delete(packet);
5484 rule_modify_actions(struct rule *rule_)
5486 struct rule_dpif *rule = rule_dpif_cast(rule_);
5488 complete_operation(rule);
5491 /* Sends 'packet' out 'ofport'.
5492 * May modify 'packet'.
5493 * Returns 0 if successful, otherwise a positive errno value. */
5495 send_packet(const struct ofport_dpif *ofport, struct ofpbuf *packet)
5497 const struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5498 uint64_t odp_actions_stub[1024 / 8];
5499 struct ofpbuf key, odp_actions;
5500 struct odputil_keybuf keybuf;
5505 flow_extract(packet, 0, 0, NULL, OFPP_LOCAL, &flow);
5506 if (netdev_vport_is_patch(ofport->up.netdev)) {
5507 struct ofproto_dpif *peer_ofproto;
5508 struct dpif_flow_stats stats;
5509 struct ofport_dpif *peer;
5510 struct rule_dpif *rule;
5512 peer = ofport_get_peer(ofport);
5517 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5518 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5519 netdev_vport_inc_rx(peer->up.netdev, &stats);
5521 flow.in_port = peer->up.ofp_port;
5522 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5523 rule = rule_dpif_lookup(peer_ofproto, &flow);
5524 rule_dpif_execute(rule, &flow, packet);
5529 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
5531 if (ofport->tnl_port) {
5532 struct dpif_flow_stats stats;
5534 odp_port = tnl_port_send(ofport->tnl_port, &flow);
5535 if (odp_port == OVSP_NONE) {
5539 dpif_flow_stats_extract(&flow, packet, time_msec(), &stats);
5540 netdev_vport_inc_tx(ofport->up.netdev, &stats);
5541 odp_put_tunnel_action(&flow.tunnel, &odp_actions);
5542 odp_put_skb_mark_action(flow.skb_mark, &odp_actions);
5544 odp_port = vsp_realdev_to_vlandev(ofproto, ofport->odp_port,
5546 if (odp_port != ofport->odp_port) {
5547 eth_pop_vlan(packet);
5548 flow.vlan_tci = htons(0);
5552 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
5553 odp_flow_key_from_flow(&key, &flow,
5554 ofp_port_to_odp_port(ofproto, flow.in_port));
5556 compose_sflow_action(ofproto, &odp_actions, &flow, odp_port);
5558 nl_msg_put_u32(&odp_actions, OVS_ACTION_ATTR_OUTPUT, odp_port);
5559 error = dpif_execute(ofproto->backer->dpif,
5561 odp_actions.data, odp_actions.size,
5563 ofpbuf_uninit(&odp_actions);
5566 VLOG_WARN_RL(&rl, "%s: failed to send packet on port %"PRIu32" (%s)",
5567 ofproto->up.name, odp_port, strerror(error));
5569 ofproto_update_local_port_stats(ofport->up.ofproto, packet->size, 0);
5573 /* OpenFlow to datapath action translation. */
5575 static bool may_receive(const struct ofport_dpif *, struct action_xlate_ctx *);
5576 static void do_xlate_actions(const struct ofpact *, size_t ofpacts_len,
5577 struct action_xlate_ctx *);
5578 static void xlate_normal(struct action_xlate_ctx *);
5580 /* Composes an ODP action for a "slow path" action for 'flow' within 'ofproto'.
5581 * The action will state 'slow' as the reason that the action is in the slow
5582 * path. (This is purely informational: it allows a human viewing "ovs-dpctl
5583 * dump-flows" output to see why a flow is in the slow path.)
5585 * The 'stub_size' bytes in 'stub' will be used to store the action.
5586 * 'stub_size' must be large enough for the action.
5588 * The action and its size will be stored in '*actionsp' and '*actions_lenp',
5591 compose_slow_path(const struct ofproto_dpif *ofproto, const struct flow *flow,
5592 enum slow_path_reason slow,
5593 uint64_t *stub, size_t stub_size,
5594 const struct nlattr **actionsp, size_t *actions_lenp)
5596 union user_action_cookie cookie;
5599 cookie.type = USER_ACTION_COOKIE_SLOW_PATH;
5600 cookie.slow_path.unused = 0;
5601 cookie.slow_path.reason = slow;
5603 ofpbuf_use_stack(&buf, stub, stub_size);
5604 if (slow & (SLOW_CFM | SLOW_LACP | SLOW_STP)) {
5605 uint32_t pid = dpif_port_get_pid(ofproto->backer->dpif, UINT32_MAX);
5606 odp_put_userspace_action(pid, &cookie, sizeof cookie, &buf);
5608 put_userspace_action(ofproto, &buf, flow, &cookie);
5610 *actionsp = buf.data;
5611 *actions_lenp = buf.size;
5615 put_userspace_action(const struct ofproto_dpif *ofproto,
5616 struct ofpbuf *odp_actions,
5617 const struct flow *flow,
5618 const union user_action_cookie *cookie)
5622 pid = dpif_port_get_pid(ofproto->backer->dpif,
5623 ofp_port_to_odp_port(ofproto, flow->in_port));
5625 return odp_put_userspace_action(pid, cookie, sizeof *cookie, odp_actions);
5629 compose_sflow_cookie(const struct ofproto_dpif *ofproto,
5630 ovs_be16 vlan_tci, uint32_t odp_port,
5631 unsigned int n_outputs, union user_action_cookie *cookie)
5635 cookie->type = USER_ACTION_COOKIE_SFLOW;
5636 cookie->sflow.vlan_tci = vlan_tci;
5638 /* See http://www.sflow.org/sflow_version_5.txt (search for "Input/output
5639 * port information") for the interpretation of cookie->output. */
5640 switch (n_outputs) {
5642 /* 0x40000000 | 256 means "packet dropped for unknown reason". */
5643 cookie->sflow.output = 0x40000000 | 256;
5647 ifindex = dpif_sflow_odp_port_to_ifindex(ofproto->sflow, odp_port);
5649 cookie->sflow.output = ifindex;
5654 /* 0x80000000 means "multiple output ports. */
5655 cookie->sflow.output = 0x80000000 | n_outputs;
5660 /* Compose SAMPLE action for sFlow. */
5662 compose_sflow_action(const struct ofproto_dpif *ofproto,
5663 struct ofpbuf *odp_actions,
5664 const struct flow *flow,
5667 uint32_t probability;
5668 union user_action_cookie cookie;
5669 size_t sample_offset, actions_offset;
5672 if (!ofproto->sflow || flow->in_port == OFPP_NONE) {
5676 sample_offset = nl_msg_start_nested(odp_actions, OVS_ACTION_ATTR_SAMPLE);
5678 /* Number of packets out of UINT_MAX to sample. */
5679 probability = dpif_sflow_get_probability(ofproto->sflow);
5680 nl_msg_put_u32(odp_actions, OVS_SAMPLE_ATTR_PROBABILITY, probability);
5682 actions_offset = nl_msg_start_nested(odp_actions, OVS_SAMPLE_ATTR_ACTIONS);
5683 compose_sflow_cookie(ofproto, htons(0), odp_port,
5684 odp_port == OVSP_NONE ? 0 : 1, &cookie);
5685 cookie_offset = put_userspace_action(ofproto, odp_actions, flow, &cookie);
5687 nl_msg_end_nested(odp_actions, actions_offset);
5688 nl_msg_end_nested(odp_actions, sample_offset);
5689 return cookie_offset;
5692 /* SAMPLE action must be first action in any given list of actions.
5693 * At this point we do not have all information required to build it. So try to
5694 * build sample action as complete as possible. */
5696 add_sflow_action(struct action_xlate_ctx *ctx)
5698 ctx->user_cookie_offset = compose_sflow_action(ctx->ofproto,
5700 &ctx->flow, OVSP_NONE);
5701 ctx->sflow_odp_port = 0;
5702 ctx->sflow_n_outputs = 0;
5705 /* Fix SAMPLE action according to data collected while composing ODP actions.
5706 * We need to fix SAMPLE actions OVS_SAMPLE_ATTR_ACTIONS attribute, i.e. nested
5707 * USERSPACE action's user-cookie which is required for sflow. */
5709 fix_sflow_action(struct action_xlate_ctx *ctx)
5711 const struct flow *base = &ctx->base_flow;
5712 union user_action_cookie *cookie;
5714 if (!ctx->user_cookie_offset) {
5718 cookie = ofpbuf_at(ctx->odp_actions, ctx->user_cookie_offset,
5720 ovs_assert(cookie->type == USER_ACTION_COOKIE_SFLOW);
5722 compose_sflow_cookie(ctx->ofproto, base->vlan_tci,
5723 ctx->sflow_odp_port, ctx->sflow_n_outputs, cookie);
5727 compose_output_action__(struct action_xlate_ctx *ctx, uint16_t ofp_port,
5730 const struct ofport_dpif *ofport = get_ofp_port(ctx->ofproto, ofp_port);
5731 ovs_be16 flow_vlan_tci = ctx->flow.vlan_tci;
5732 ovs_be64 flow_tun_id = ctx->flow.tunnel.tun_id;
5733 uint8_t flow_nw_tos = ctx->flow.nw_tos;
5734 struct priority_to_dscp *pdscp;
5735 uint32_t out_port, odp_port;
5737 /* If 'struct flow' gets additional metadata, we'll need to zero it out
5738 * before traversing a patch port. */
5739 BUILD_ASSERT_DECL(FLOW_WC_SEQ == 19);
5742 xlate_report(ctx, "Nonexistent output port");
5744 } else if (ofport->up.pp.config & OFPUTIL_PC_NO_FWD) {
5745 xlate_report(ctx, "OFPPC_NO_FWD set, skipping output");
5747 } else if (check_stp && !stp_forward_in_state(ofport->stp_state)) {
5748 xlate_report(ctx, "STP not in forwarding state, skipping output");
5752 if (netdev_vport_is_patch(ofport->up.netdev)) {
5753 struct ofport_dpif *peer = ofport_get_peer(ofport);
5754 struct flow old_flow = ctx->flow;
5755 const struct ofproto_dpif *peer_ofproto;
5756 enum slow_path_reason special;
5757 struct ofport_dpif *in_port;
5760 xlate_report(ctx, "Nonexistent patch port peer");
5764 peer_ofproto = ofproto_dpif_cast(peer->up.ofproto);
5765 if (peer_ofproto->backer != ctx->ofproto->backer) {
5766 xlate_report(ctx, "Patch port peer on a different datapath");
5770 ctx->ofproto = ofproto_dpif_cast(peer->up.ofproto);
5771 ctx->flow.in_port = peer->up.ofp_port;
5772 ctx->flow.metadata = htonll(0);
5773 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
5774 memset(ctx->flow.regs, 0, sizeof ctx->flow.regs);
5776 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
5777 special = process_special(ctx->ofproto, &ctx->flow, in_port,
5780 ctx->slow |= special;
5781 } else if (!in_port || may_receive(in_port, ctx)) {
5782 if (!in_port || stp_forward_in_state(in_port->stp_state)) {
5783 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
5785 /* Forwarding is disabled by STP. Let OFPP_NORMAL and the
5786 * learning action look at the packet, then drop it. */
5787 struct flow old_base_flow = ctx->base_flow;
5788 size_t old_size = ctx->odp_actions->size;
5789 xlate_table_action(ctx, ctx->flow.in_port, 0, true);
5790 ctx->base_flow = old_base_flow;
5791 ctx->odp_actions->size = old_size;
5795 ctx->flow = old_flow;
5796 ctx->ofproto = ofproto_dpif_cast(ofport->up.ofproto);
5798 if (ctx->resubmit_stats) {
5799 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
5800 netdev_vport_inc_rx(peer->up.netdev, ctx->resubmit_stats);
5806 pdscp = get_priority(ofport, ctx->flow.skb_priority);
5808 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
5809 ctx->flow.nw_tos |= pdscp->dscp;
5812 if (ofport->tnl_port) {
5813 odp_port = tnl_port_send(ofport->tnl_port, &ctx->flow);
5814 if (odp_port == OVSP_NONE) {
5815 xlate_report(ctx, "Tunneling decided against output");
5819 if (ctx->resubmit_stats) {
5820 netdev_vport_inc_tx(ofport->up.netdev, ctx->resubmit_stats);
5822 out_port = odp_port;
5823 commit_odp_tunnel_action(&ctx->flow, &ctx->base_flow,
5826 odp_port = ofport->odp_port;
5827 out_port = vsp_realdev_to_vlandev(ctx->ofproto, odp_port,
5828 ctx->flow.vlan_tci);
5829 if (out_port != odp_port) {
5830 ctx->flow.vlan_tci = htons(0);
5833 commit_odp_actions(&ctx->flow, &ctx->base_flow, ctx->odp_actions);
5834 nl_msg_put_u32(ctx->odp_actions, OVS_ACTION_ATTR_OUTPUT, out_port);
5836 ctx->sflow_odp_port = odp_port;
5837 ctx->sflow_n_outputs++;
5838 ctx->nf_output_iface = ofp_port;
5839 ctx->flow.tunnel.tun_id = flow_tun_id;
5840 ctx->flow.vlan_tci = flow_vlan_tci;
5841 ctx->flow.nw_tos = flow_nw_tos;
5845 compose_output_action(struct action_xlate_ctx *ctx, uint16_t ofp_port)
5847 compose_output_action__(ctx, ofp_port, true);
5851 tag_the_flow(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
5853 struct ofproto_dpif *ofproto = ctx->ofproto;
5854 uint8_t table_id = ctx->table_id;
5856 if (table_id > 0 && table_id < N_TABLES) {
5857 struct table_dpif *table = &ofproto->tables[table_id];
5858 if (table->other_table) {
5859 ctx->tags |= (rule && rule->tag
5861 : rule_calculate_tag(&ctx->flow,
5862 &table->other_table->mask,
5868 /* Common rule processing in one place to avoid duplicating code. */
5869 static struct rule_dpif *
5870 ctx_rule_hooks(struct action_xlate_ctx *ctx, struct rule_dpif *rule,
5873 if (ctx->resubmit_hook) {
5874 ctx->resubmit_hook(ctx, rule);
5876 if (rule == NULL && may_packet_in) {
5878 * check if table configuration flags
5879 * OFPTC_TABLE_MISS_CONTROLLER, default.
5880 * OFPTC_TABLE_MISS_CONTINUE,
5881 * OFPTC_TABLE_MISS_DROP
5882 * When OF1.0, OFPTC_TABLE_MISS_CONTINUE is used. What to do?
5884 rule = rule_dpif_miss_rule(ctx->ofproto, &ctx->flow);
5886 if (rule && ctx->resubmit_stats) {
5887 rule_credit_stats(rule, ctx->resubmit_stats);
5893 xlate_table_action(struct action_xlate_ctx *ctx,
5894 uint16_t in_port, uint8_t table_id, bool may_packet_in)
5896 if (ctx->recurse < MAX_RESUBMIT_RECURSION) {
5897 struct rule_dpif *rule;
5898 uint16_t old_in_port = ctx->flow.in_port;
5899 uint8_t old_table_id = ctx->table_id;
5901 ctx->table_id = table_id;
5903 /* Look up a flow with 'in_port' as the input port. */
5904 ctx->flow.in_port = in_port;
5905 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->flow, table_id);
5907 tag_the_flow(ctx, rule);
5909 /* Restore the original input port. Otherwise OFPP_NORMAL and
5910 * OFPP_IN_PORT will have surprising behavior. */
5911 ctx->flow.in_port = old_in_port;
5913 rule = ctx_rule_hooks(ctx, rule, may_packet_in);
5916 struct rule_dpif *old_rule = ctx->rule;
5920 do_xlate_actions(rule->up.ofpacts, rule->up.ofpacts_len, ctx);
5921 ctx->rule = old_rule;
5925 ctx->table_id = old_table_id;
5927 static struct vlog_rate_limit recurse_rl = VLOG_RATE_LIMIT_INIT(1, 1);
5929 VLOG_ERR_RL(&recurse_rl, "resubmit actions recursed over %d times",
5930 MAX_RESUBMIT_RECURSION);
5931 ctx->max_resubmit_trigger = true;
5936 xlate_ofpact_resubmit(struct action_xlate_ctx *ctx,
5937 const struct ofpact_resubmit *resubmit)
5942 in_port = resubmit->in_port;
5943 if (in_port == OFPP_IN_PORT) {
5944 in_port = ctx->flow.in_port;
5947 table_id = resubmit->table_id;
5948 if (table_id == 255) {
5949 table_id = ctx->table_id;
5952 xlate_table_action(ctx, in_port, table_id, false);
5956 flood_packets(struct action_xlate_ctx *ctx, bool all)
5958 struct ofport_dpif *ofport;
5960 HMAP_FOR_EACH (ofport, up.hmap_node, &ctx->ofproto->up.ports) {
5961 uint16_t ofp_port = ofport->up.ofp_port;
5963 if (ofp_port == ctx->flow.in_port) {
5968 compose_output_action__(ctx, ofp_port, false);
5969 } else if (!(ofport->up.pp.config & OFPUTIL_PC_NO_FLOOD)) {
5970 compose_output_action(ctx, ofp_port);
5974 ctx->nf_output_iface = NF_OUT_FLOOD;
5978 execute_controller_action(struct action_xlate_ctx *ctx, int len,
5979 enum ofp_packet_in_reason reason,
5980 uint16_t controller_id)
5982 struct ofputil_packet_in pin;
5983 struct ofpbuf *packet;
5985 ctx->slow |= SLOW_CONTROLLER;
5990 packet = ofpbuf_clone(ctx->packet);
5992 if (packet->l2 && packet->l3) {
5993 struct eth_header *eh;
5994 uint16_t mpls_depth;
5996 eth_pop_vlan(packet);
5999 memcpy(eh->eth_src, ctx->flow.dl_src, sizeof eh->eth_src);
6000 memcpy(eh->eth_dst, ctx->flow.dl_dst, sizeof eh->eth_dst);
6002 if (ctx->flow.vlan_tci & htons(VLAN_CFI)) {
6003 eth_push_vlan(packet, ctx->flow.vlan_tci);
6006 mpls_depth = eth_mpls_depth(packet);
6008 if (mpls_depth < ctx->flow.mpls_depth) {
6009 push_mpls(packet, ctx->flow.dl_type, ctx->flow.mpls_lse);
6010 } else if (mpls_depth > ctx->flow.mpls_depth) {
6011 pop_mpls(packet, ctx->flow.dl_type);
6012 } else if (mpls_depth) {
6013 set_mpls_lse(packet, ctx->flow.mpls_lse);
6017 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6018 packet_set_ipv4(packet, ctx->flow.nw_src, ctx->flow.nw_dst,
6019 ctx->flow.nw_tos, ctx->flow.nw_ttl);
6023 if (ctx->flow.nw_proto == IPPROTO_TCP) {
6024 packet_set_tcp_port(packet, ctx->flow.tp_src,
6026 } else if (ctx->flow.nw_proto == IPPROTO_UDP) {
6027 packet_set_udp_port(packet, ctx->flow.tp_src,
6034 pin.packet = packet->data;
6035 pin.packet_len = packet->size;
6036 pin.reason = reason;
6037 pin.controller_id = controller_id;
6038 pin.table_id = ctx->table_id;
6039 pin.cookie = ctx->rule ? ctx->rule->up.flow_cookie : 0;
6042 flow_get_metadata(&ctx->flow, &pin.fmd);
6044 connmgr_send_packet_in(ctx->ofproto->up.connmgr, &pin);
6045 ofpbuf_delete(packet);
6049 execute_mpls_push_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type)
6051 ovs_assert(eth_type_mpls(eth_type));
6053 if (ctx->base_flow.mpls_depth) {
6054 ctx->flow.mpls_lse &= ~htonl(MPLS_BOS_MASK);
6055 ctx->flow.mpls_depth++;
6060 if (ctx->flow.dl_type == htons(ETH_TYPE_IPV6)) {
6061 label = htonl(0x2); /* IPV6 Explicit Null. */
6063 label = htonl(0x0); /* IPV4 Explicit Null. */
6065 tc = (ctx->flow.nw_tos & IP_DSCP_MASK) >> 2;
6066 ttl = ctx->flow.nw_ttl ? ctx->flow.nw_ttl : 0x40;
6067 ctx->flow.mpls_lse = set_mpls_lse_values(ttl, tc, 1, label);
6068 ctx->flow.encap_dl_type = ctx->flow.dl_type;
6069 ctx->flow.mpls_depth = 1;
6071 ctx->flow.dl_type = eth_type;
6075 execute_mpls_pop_action(struct action_xlate_ctx *ctx, ovs_be16 eth_type)
6077 ovs_assert(eth_type_mpls(ctx->flow.dl_type));
6078 ovs_assert(!eth_type_mpls(eth_type));
6080 if (ctx->flow.mpls_depth) {
6081 ctx->flow.mpls_depth--;
6082 ctx->flow.mpls_lse = htonl(0);
6083 if (!ctx->flow.mpls_depth) {
6084 ctx->flow.dl_type = eth_type;
6085 ctx->flow.encap_dl_type = htons(0);
6091 compose_dec_ttl(struct action_xlate_ctx *ctx, struct ofpact_cnt_ids *ids)
6093 if (ctx->flow.dl_type != htons(ETH_TYPE_IP) &&
6094 ctx->flow.dl_type != htons(ETH_TYPE_IPV6)) {
6098 if (ctx->flow.nw_ttl > 1) {
6104 for (i = 0; i < ids->n_controllers; i++) {
6105 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL,
6109 /* Stop processing for current table. */
6115 execute_set_mpls_ttl_action(struct action_xlate_ctx *ctx, uint8_t ttl)
6117 if (!eth_type_mpls(ctx->flow.dl_type)) {
6121 set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl);
6126 execute_dec_mpls_ttl_action(struct action_xlate_ctx *ctx)
6128 uint8_t ttl = mpls_lse_to_ttl(ctx->flow.mpls_lse);
6130 if (!eth_type_mpls(ctx->flow.dl_type)) {
6136 set_mpls_lse_ttl(&ctx->flow.mpls_lse, ttl);
6139 execute_controller_action(ctx, UINT16_MAX, OFPR_INVALID_TTL, 0);
6141 /* Stop processing for current table. */
6147 xlate_output_action(struct action_xlate_ctx *ctx,
6148 uint16_t port, uint16_t max_len, bool may_packet_in)
6150 uint16_t prev_nf_output_iface = ctx->nf_output_iface;
6152 ctx->nf_output_iface = NF_OUT_DROP;
6156 compose_output_action(ctx, ctx->flow.in_port);
6159 xlate_table_action(ctx, ctx->flow.in_port, 0, may_packet_in);
6165 flood_packets(ctx, false);
6168 flood_packets(ctx, true);
6170 case OFPP_CONTROLLER:
6171 execute_controller_action(ctx, max_len, OFPR_ACTION, 0);
6177 if (port != ctx->flow.in_port) {
6178 compose_output_action(ctx, port);
6180 xlate_report(ctx, "skipping output to input port");
6185 if (prev_nf_output_iface == NF_OUT_FLOOD) {
6186 ctx->nf_output_iface = NF_OUT_FLOOD;
6187 } else if (ctx->nf_output_iface == NF_OUT_DROP) {
6188 ctx->nf_output_iface = prev_nf_output_iface;
6189 } else if (prev_nf_output_iface != NF_OUT_DROP &&
6190 ctx->nf_output_iface != NF_OUT_FLOOD) {
6191 ctx->nf_output_iface = NF_OUT_MULTI;
6196 xlate_output_reg_action(struct action_xlate_ctx *ctx,
6197 const struct ofpact_output_reg *or)
6199 uint64_t port = mf_get_subfield(&or->src, &ctx->flow);
6200 if (port <= UINT16_MAX) {
6201 xlate_output_action(ctx, port, or->max_len, false);
6206 xlate_enqueue_action(struct action_xlate_ctx *ctx,
6207 const struct ofpact_enqueue *enqueue)
6209 uint16_t ofp_port = enqueue->port;
6210 uint32_t queue_id = enqueue->queue;
6211 uint32_t flow_priority, priority;
6214 /* Translate queue to priority. */
6215 error = dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6216 queue_id, &priority);
6218 /* Fall back to ordinary output action. */
6219 xlate_output_action(ctx, enqueue->port, 0, false);
6223 /* Check output port. */
6224 if (ofp_port == OFPP_IN_PORT) {
6225 ofp_port = ctx->flow.in_port;
6226 } else if (ofp_port == ctx->flow.in_port) {
6230 /* Add datapath actions. */
6231 flow_priority = ctx->flow.skb_priority;
6232 ctx->flow.skb_priority = priority;
6233 compose_output_action(ctx, ofp_port);
6234 ctx->flow.skb_priority = flow_priority;
6236 /* Update NetFlow output port. */
6237 if (ctx->nf_output_iface == NF_OUT_DROP) {
6238 ctx->nf_output_iface = ofp_port;
6239 } else if (ctx->nf_output_iface != NF_OUT_FLOOD) {
6240 ctx->nf_output_iface = NF_OUT_MULTI;
6245 xlate_set_queue_action(struct action_xlate_ctx *ctx, uint32_t queue_id)
6247 uint32_t skb_priority;
6249 if (!dpif_queue_to_priority(ctx->ofproto->backer->dpif,
6250 queue_id, &skb_priority)) {
6251 ctx->flow.skb_priority = skb_priority;
6253 /* Couldn't translate queue to a priority. Nothing to do. A warning
6254 * has already been logged. */
6258 struct xlate_reg_state {
6264 slave_enabled_cb(uint16_t ofp_port, void *ofproto_)
6266 struct ofproto_dpif *ofproto = ofproto_;
6267 struct ofport_dpif *port;
6277 case OFPP_CONTROLLER: /* Not supported by the bundle action. */
6280 port = get_ofp_port(ofproto, ofp_port);
6281 return port ? port->may_enable : false;
6286 xlate_bundle_action(struct action_xlate_ctx *ctx,
6287 const struct ofpact_bundle *bundle)
6291 port = bundle_execute(bundle, &ctx->flow, slave_enabled_cb, ctx->ofproto);
6292 if (bundle->dst.field) {
6293 nxm_reg_load(&bundle->dst, port, &ctx->flow);
6295 xlate_output_action(ctx, port, 0, false);
6300 xlate_learn_action(struct action_xlate_ctx *ctx,
6301 const struct ofpact_learn *learn)
6303 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(5, 1);
6304 struct ofputil_flow_mod fm;
6305 uint64_t ofpacts_stub[1024 / 8];
6306 struct ofpbuf ofpacts;
6309 ofpbuf_use_stack(&ofpacts, ofpacts_stub, sizeof ofpacts_stub);
6310 learn_execute(learn, &ctx->flow, &fm, &ofpacts);
6312 error = ofproto_flow_mod(&ctx->ofproto->up, &fm);
6313 if (error && !VLOG_DROP_WARN(&rl)) {
6314 VLOG_WARN("learning action failed to modify flow table (%s)",
6315 ofperr_get_name(error));
6318 ofpbuf_uninit(&ofpacts);
6321 /* Reduces '*timeout' to no more than 'max'. A value of zero in either case
6322 * means "infinite". */
6324 reduce_timeout(uint16_t max, uint16_t *timeout)
6326 if (max && (!*timeout || *timeout > max)) {
6332 xlate_fin_timeout(struct action_xlate_ctx *ctx,
6333 const struct ofpact_fin_timeout *oft)
6335 if (ctx->tcp_flags & (TCP_FIN | TCP_RST) && ctx->rule) {
6336 struct rule_dpif *rule = ctx->rule;
6338 reduce_timeout(oft->fin_idle_timeout, &rule->up.idle_timeout);
6339 reduce_timeout(oft->fin_hard_timeout, &rule->up.hard_timeout);
6344 may_receive(const struct ofport_dpif *port, struct action_xlate_ctx *ctx)
6346 if (port->up.pp.config & (eth_addr_equals(ctx->flow.dl_dst, eth_addr_stp)
6347 ? OFPUTIL_PC_NO_RECV_STP
6348 : OFPUTIL_PC_NO_RECV)) {
6352 /* Only drop packets here if both forwarding and learning are
6353 * disabled. If just learning is enabled, we need to have
6354 * OFPP_NORMAL and the learning action have a look at the packet
6355 * before we can drop it. */
6356 if (!stp_forward_in_state(port->stp_state)
6357 && !stp_learn_in_state(port->stp_state)) {
6365 tunnel_ecn_ok(struct action_xlate_ctx *ctx)
6367 if (is_ip_any(&ctx->base_flow)
6368 && (ctx->base_flow.tunnel.ip_tos & IP_ECN_MASK) == IP_ECN_CE) {
6369 if ((ctx->base_flow.nw_tos & IP_ECN_MASK) == IP_ECN_NOT_ECT) {
6370 VLOG_WARN_RL(&rl, "dropping tunnel packet marked ECN CE"
6371 " but is not ECN capable");
6374 /* Set the ECN CE value in the tunneled packet. */
6375 ctx->flow.nw_tos |= IP_ECN_CE;
6383 do_xlate_actions(const struct ofpact *ofpacts, size_t ofpacts_len,
6384 struct action_xlate_ctx *ctx)
6386 bool was_evictable = true;
6387 const struct ofpact *a;
6390 /* Don't let the rule we're working on get evicted underneath us. */
6391 was_evictable = ctx->rule->up.evictable;
6392 ctx->rule->up.evictable = false;
6395 do_xlate_actions_again:
6396 OFPACT_FOR_EACH (a, ofpacts, ofpacts_len) {
6397 struct ofpact_controller *controller;
6398 const struct ofpact_metadata *metadata;
6406 xlate_output_action(ctx, ofpact_get_OUTPUT(a)->port,
6407 ofpact_get_OUTPUT(a)->max_len, true);
6410 case OFPACT_CONTROLLER:
6411 controller = ofpact_get_CONTROLLER(a);
6412 execute_controller_action(ctx, controller->max_len,
6414 controller->controller_id);
6417 case OFPACT_ENQUEUE:
6418 xlate_enqueue_action(ctx, ofpact_get_ENQUEUE(a));
6421 case OFPACT_SET_VLAN_VID:
6422 ctx->flow.vlan_tci &= ~htons(VLAN_VID_MASK);
6423 ctx->flow.vlan_tci |= (htons(ofpact_get_SET_VLAN_VID(a)->vlan_vid)
6427 case OFPACT_SET_VLAN_PCP:
6428 ctx->flow.vlan_tci &= ~htons(VLAN_PCP_MASK);
6429 ctx->flow.vlan_tci |= htons((ofpact_get_SET_VLAN_PCP(a)->vlan_pcp
6434 case OFPACT_STRIP_VLAN:
6435 ctx->flow.vlan_tci = htons(0);
6438 case OFPACT_PUSH_VLAN:
6439 /* XXX 802.1AD(QinQ) */
6440 ctx->flow.vlan_tci = htons(VLAN_CFI);
6443 case OFPACT_SET_ETH_SRC:
6444 memcpy(ctx->flow.dl_src, ofpact_get_SET_ETH_SRC(a)->mac,
6448 case OFPACT_SET_ETH_DST:
6449 memcpy(ctx->flow.dl_dst, ofpact_get_SET_ETH_DST(a)->mac,
6453 case OFPACT_SET_IPV4_SRC:
6454 ctx->flow.nw_src = ofpact_get_SET_IPV4_SRC(a)->ipv4;
6457 case OFPACT_SET_IPV4_DST:
6458 ctx->flow.nw_dst = ofpact_get_SET_IPV4_DST(a)->ipv4;
6461 case OFPACT_SET_IPV4_DSCP:
6462 /* OpenFlow 1.0 only supports IPv4. */
6463 if (ctx->flow.dl_type == htons(ETH_TYPE_IP)) {
6464 ctx->flow.nw_tos &= ~IP_DSCP_MASK;
6465 ctx->flow.nw_tos |= ofpact_get_SET_IPV4_DSCP(a)->dscp;
6469 case OFPACT_SET_L4_SRC_PORT:
6470 ctx->flow.tp_src = htons(ofpact_get_SET_L4_SRC_PORT(a)->port);
6473 case OFPACT_SET_L4_DST_PORT:
6474 ctx->flow.tp_dst = htons(ofpact_get_SET_L4_DST_PORT(a)->port);
6477 case OFPACT_RESUBMIT:
6478 xlate_ofpact_resubmit(ctx, ofpact_get_RESUBMIT(a));
6481 case OFPACT_SET_TUNNEL:
6482 ctx->flow.tunnel.tun_id = htonll(ofpact_get_SET_TUNNEL(a)->tun_id);
6485 case OFPACT_SET_QUEUE:
6486 xlate_set_queue_action(ctx, ofpact_get_SET_QUEUE(a)->queue_id);
6489 case OFPACT_POP_QUEUE:
6490 ctx->flow.skb_priority = ctx->orig_skb_priority;
6493 case OFPACT_REG_MOVE:
6494 nxm_execute_reg_move(ofpact_get_REG_MOVE(a), &ctx->flow);
6497 case OFPACT_REG_LOAD:
6498 nxm_execute_reg_load(ofpact_get_REG_LOAD(a), &ctx->flow);
6501 case OFPACT_STACK_PUSH:
6502 nxm_execute_stack_push(ofpact_get_STACK_PUSH(a), &ctx->flow,
6506 case OFPACT_STACK_POP:
6507 nxm_execute_stack_pop(ofpact_get_STACK_POP(a), &ctx->flow,
6511 case OFPACT_PUSH_MPLS:
6512 execute_mpls_push_action(ctx, ofpact_get_PUSH_MPLS(a)->ethertype);
6515 case OFPACT_POP_MPLS:
6516 execute_mpls_pop_action(ctx, ofpact_get_POP_MPLS(a)->ethertype);
6519 case OFPACT_SET_MPLS_TTL:
6520 if (execute_set_mpls_ttl_action(ctx, ofpact_get_SET_MPLS_TTL(a)->ttl)) {
6525 case OFPACT_DEC_MPLS_TTL:
6526 if (execute_dec_mpls_ttl_action(ctx)) {
6531 case OFPACT_DEC_TTL:
6532 if (compose_dec_ttl(ctx, ofpact_get_DEC_TTL(a))) {
6538 /* Nothing to do. */
6541 case OFPACT_MULTIPATH:
6542 multipath_execute(ofpact_get_MULTIPATH(a), &ctx->flow);
6546 ctx->ofproto->has_bundle_action = true;
6547 xlate_bundle_action(ctx, ofpact_get_BUNDLE(a));
6550 case OFPACT_OUTPUT_REG:
6551 xlate_output_reg_action(ctx, ofpact_get_OUTPUT_REG(a));
6555 ctx->has_learn = true;
6556 if (ctx->may_learn) {
6557 xlate_learn_action(ctx, ofpact_get_LEARN(a));
6565 case OFPACT_FIN_TIMEOUT:
6566 ctx->has_fin_timeout = true;
6567 xlate_fin_timeout(ctx, ofpact_get_FIN_TIMEOUT(a));
6570 case OFPACT_CLEAR_ACTIONS:
6572 * Nothing to do because writa-actions is not supported for now.
6573 * When writa-actions is supported, clear-actions also must
6574 * be supported at the same time.
6578 case OFPACT_WRITE_METADATA:
6579 metadata = ofpact_get_WRITE_METADATA(a);
6580 ctx->flow.metadata &= ~metadata->mask;
6581 ctx->flow.metadata |= metadata->metadata & metadata->mask;
6584 case OFPACT_GOTO_TABLE: {
6585 /* It is assumed that goto-table is the last action. */
6586 struct ofpact_goto_table *ogt = ofpact_get_GOTO_TABLE(a);
6587 struct rule_dpif *rule;
6589 ovs_assert(ctx->table_id < ogt->table_id);
6591 ctx->table_id = ogt->table_id;
6593 /* Look up a flow from the new table. */
6594 rule = rule_dpif_lookup__(ctx->ofproto, &ctx->flow, ctx->table_id);
6596 tag_the_flow(ctx, rule);
6598 rule = ctx_rule_hooks(ctx, rule, true);
6602 ctx->rule->up.evictable = was_evictable;
6605 was_evictable = rule->up.evictable;
6606 rule->up.evictable = false;
6608 /* Tail recursion removal. */
6609 ofpacts = rule->up.ofpacts;
6610 ofpacts_len = rule->up.ofpacts_len;
6611 goto do_xlate_actions_again;
6620 ctx->rule->up.evictable = was_evictable;
6625 action_xlate_ctx_init(struct action_xlate_ctx *ctx,
6626 struct ofproto_dpif *ofproto, const struct flow *flow,
6627 const struct initial_vals *initial_vals,
6628 struct rule_dpif *rule,
6629 uint8_t tcp_flags, const struct ofpbuf *packet)
6631 ovs_be64 initial_tun_id = flow->tunnel.tun_id;
6633 /* Flow initialization rules:
6634 * - 'base_flow' must match the kernel's view of the packet at the
6635 * time that action processing starts. 'flow' represents any
6636 * transformations we wish to make through actions.
6637 * - By default 'base_flow' and 'flow' are the same since the input
6638 * packet matches the output before any actions are applied.
6639 * - When using VLAN splinters, 'base_flow''s VLAN is set to the value
6640 * of the received packet as seen by the kernel. If we later output
6641 * to another device without any modifications this will cause us to
6642 * insert a new tag since the original one was stripped off by the
6644 * - Tunnel 'flow' is largely cleared when transitioning between
6645 * the input and output stages since it does not make sense to output
6646 * a packet with the exact headers that it was received with (i.e.
6647 * the destination IP is us). The one exception is the tun_id, which
6648 * is preserved to allow use in later resubmit lookups and loads into
6650 * - Tunnel 'base_flow' is completely cleared since that is what the
6651 * kernel does. If we wish to maintain the original values an action
6652 * needs to be generated. */
6654 ctx->ofproto = ofproto;
6656 memset(&ctx->flow.tunnel, 0, sizeof ctx->flow.tunnel);
6657 ctx->base_flow = ctx->flow;
6658 ctx->base_flow.vlan_tci = initial_vals->vlan_tci;
6659 ctx->base_flow.tunnel.ip_tos = initial_vals->tunnel_ip_tos;
6660 ctx->flow.tunnel.tun_id = initial_tun_id;
6662 ctx->packet = packet;
6663 ctx->may_learn = packet != NULL;
6664 ctx->tcp_flags = tcp_flags;
6665 ctx->resubmit_hook = NULL;
6666 ctx->report_hook = NULL;
6667 ctx->resubmit_stats = NULL;
6670 /* Translates the 'ofpacts_len' bytes of "struct ofpacts" starting at 'ofpacts'
6671 * into datapath actions in 'odp_actions', using 'ctx'. */
6673 xlate_actions(struct action_xlate_ctx *ctx,
6674 const struct ofpact *ofpacts, size_t ofpacts_len,
6675 struct ofpbuf *odp_actions)
6677 /* Normally false. Set to true if we ever hit MAX_RESUBMIT_RECURSION, so
6678 * that in the future we always keep a copy of the original flow for
6679 * tracing purposes. */
6680 static bool hit_resubmit_limit;
6682 enum slow_path_reason special;
6683 struct ofport_dpif *in_port;
6684 struct flow orig_flow;
6686 COVERAGE_INC(ofproto_dpif_xlate);
6688 ofpbuf_clear(odp_actions);
6689 ofpbuf_reserve(odp_actions, NL_A_U32_SIZE);
6691 ctx->odp_actions = odp_actions;
6694 ctx->has_learn = false;
6695 ctx->has_normal = false;
6696 ctx->has_fin_timeout = false;
6697 ctx->nf_output_iface = NF_OUT_DROP;
6700 ctx->max_resubmit_trigger = false;
6701 ctx->orig_skb_priority = ctx->flow.skb_priority;
6705 ofpbuf_use_stub(&ctx->stack, ctx->init_stack, sizeof ctx->init_stack);
6707 if (ctx->ofproto->has_mirrors || hit_resubmit_limit) {
6708 /* Do this conditionally because the copy is expensive enough that it
6709 * shows up in profiles. */
6710 orig_flow = ctx->flow;
6713 if (ctx->flow.nw_frag & FLOW_NW_FRAG_ANY) {
6714 switch (ctx->ofproto->up.frag_handling) {
6715 case OFPC_FRAG_NORMAL:
6716 /* We must pretend that transport ports are unavailable. */
6717 ctx->flow.tp_src = ctx->base_flow.tp_src = htons(0);
6718 ctx->flow.tp_dst = ctx->base_flow.tp_dst = htons(0);
6721 case OFPC_FRAG_DROP:
6724 case OFPC_FRAG_REASM:
6727 case OFPC_FRAG_NX_MATCH:
6728 /* Nothing to do. */
6731 case OFPC_INVALID_TTL_TO_CONTROLLER:
6736 in_port = get_ofp_port(ctx->ofproto, ctx->flow.in_port);
6737 special = process_special(ctx->ofproto, &ctx->flow, in_port, ctx->packet);
6739 ctx->slow |= special;
6741 static struct vlog_rate_limit trace_rl = VLOG_RATE_LIMIT_INIT(1, 1);
6742 struct initial_vals initial_vals;
6743 uint32_t local_odp_port;
6745 initial_vals.vlan_tci = ctx->base_flow.vlan_tci;
6746 initial_vals.tunnel_ip_tos = ctx->base_flow.tunnel.ip_tos;
6748 add_sflow_action(ctx);
6750 if (tunnel_ecn_ok(ctx) && (!in_port || may_receive(in_port, ctx))) {
6751 do_xlate_actions(ofpacts, ofpacts_len, ctx);
6753 /* We've let OFPP_NORMAL and the learning action look at the
6754 * packet, so drop it now if forwarding is disabled. */
6755 if (in_port && !stp_forward_in_state(in_port->stp_state)) {
6756 ofpbuf_clear(ctx->odp_actions);
6757 add_sflow_action(ctx);
6761 if (ctx->max_resubmit_trigger && !ctx->resubmit_hook) {
6762 if (!hit_resubmit_limit) {
6763 /* We didn't record the original flow. Make sure we do from
6765 hit_resubmit_limit = true;
6766 } else if (!VLOG_DROP_ERR(&trace_rl)) {
6767 struct ds ds = DS_EMPTY_INITIALIZER;
6769 ofproto_trace(ctx->ofproto, &orig_flow, ctx->packet,
6770 &initial_vals, &ds);
6771 VLOG_ERR("Trace triggered by excessive resubmit "
6772 "recursion:\n%s", ds_cstr(&ds));
6777 local_odp_port = ofp_port_to_odp_port(ctx->ofproto, OFPP_LOCAL);
6778 if (!connmgr_may_set_up_flow(ctx->ofproto->up.connmgr, &ctx->flow,
6780 ctx->odp_actions->data,
6781 ctx->odp_actions->size)) {
6782 ctx->slow |= SLOW_IN_BAND;
6784 && connmgr_msg_in_hook(ctx->ofproto->up.connmgr, &ctx->flow,
6786 compose_output_action(ctx, OFPP_LOCAL);
6789 if (ctx->ofproto->has_mirrors) {
6790 add_mirror_actions(ctx, &orig_flow);
6792 fix_sflow_action(ctx);
6795 ofpbuf_uninit(&ctx->stack);
6798 /* Translates the 'ofpacts_len' bytes of "struct ofpact"s starting at 'ofpacts'
6799 * into datapath actions, using 'ctx', and discards the datapath actions. */
6801 xlate_actions_for_side_effects(struct action_xlate_ctx *ctx,
6802 const struct ofpact *ofpacts,
6805 uint64_t odp_actions_stub[1024 / 8];
6806 struct ofpbuf odp_actions;
6808 ofpbuf_use_stub(&odp_actions, odp_actions_stub, sizeof odp_actions_stub);
6809 xlate_actions(ctx, ofpacts, ofpacts_len, &odp_actions);
6810 ofpbuf_uninit(&odp_actions);
6814 xlate_report(struct action_xlate_ctx *ctx, const char *s)
6816 if (ctx->report_hook) {
6817 ctx->report_hook(ctx, s);
6821 /* OFPP_NORMAL implementation. */
6823 static struct ofport_dpif *ofbundle_get_a_port(const struct ofbundle *);
6825 /* Given 'vid', the VID obtained from the 802.1Q header that was received as
6826 * part of a packet (specify 0 if there was no 802.1Q header), and 'in_bundle',
6827 * the bundle on which the packet was received, returns the VLAN to which the
6830 * Both 'vid' and the return value are in the range 0...4095. */
6832 input_vid_to_vlan(const struct ofbundle *in_bundle, uint16_t vid)
6834 switch (in_bundle->vlan_mode) {
6835 case PORT_VLAN_ACCESS:
6836 return in_bundle->vlan;
6839 case PORT_VLAN_TRUNK:
6842 case PORT_VLAN_NATIVE_UNTAGGED:
6843 case PORT_VLAN_NATIVE_TAGGED:
6844 return vid ? vid : in_bundle->vlan;
6851 /* Checks whether a packet with the given 'vid' may ingress on 'in_bundle'.
6852 * If so, returns true. Otherwise, returns false and, if 'warn' is true, logs
6855 * 'vid' should be the VID obtained from the 802.1Q header that was received as
6856 * part of a packet (specify 0 if there was no 802.1Q header), in the range
6859 input_vid_is_valid(uint16_t vid, struct ofbundle *in_bundle, bool warn)
6861 /* Allow any VID on the OFPP_NONE port. */
6862 if (in_bundle == &ofpp_none_bundle) {
6866 switch (in_bundle->vlan_mode) {
6867 case PORT_VLAN_ACCESS:
6870 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6871 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" tagged "
6872 "packet received on port %s configured as VLAN "
6873 "%"PRIu16" access port",
6874 in_bundle->ofproto->up.name, vid,
6875 in_bundle->name, in_bundle->vlan);
6881 case PORT_VLAN_NATIVE_UNTAGGED:
6882 case PORT_VLAN_NATIVE_TAGGED:
6884 /* Port must always carry its native VLAN. */
6888 case PORT_VLAN_TRUNK:
6889 if (!ofbundle_includes_vlan(in_bundle, vid)) {
6891 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
6892 VLOG_WARN_RL(&rl, "bridge %s: dropping VLAN %"PRIu16" packet "
6893 "received on port %s not configured for trunking "
6895 in_bundle->ofproto->up.name, vid,
6896 in_bundle->name, vid);
6908 /* Given 'vlan', the VLAN that a packet belongs to, and
6909 * 'out_bundle', a bundle on which the packet is to be output, returns the VID
6910 * that should be included in the 802.1Q header. (If the return value is 0,
6911 * then the 802.1Q header should only be included in the packet if there is a
6914 * Both 'vlan' and the return value are in the range 0...4095. */
6916 output_vlan_to_vid(const struct ofbundle *out_bundle, uint16_t vlan)
6918 switch (out_bundle->vlan_mode) {
6919 case PORT_VLAN_ACCESS:
6922 case PORT_VLAN_TRUNK:
6923 case PORT_VLAN_NATIVE_TAGGED:
6926 case PORT_VLAN_NATIVE_UNTAGGED:
6927 return vlan == out_bundle->vlan ? 0 : vlan;
6935 output_normal(struct action_xlate_ctx *ctx, const struct ofbundle *out_bundle,
6938 struct ofport_dpif *port;
6940 ovs_be16 tci, old_tci;
6942 vid = output_vlan_to_vid(out_bundle, vlan);
6943 if (!out_bundle->bond) {
6944 port = ofbundle_get_a_port(out_bundle);
6946 port = bond_choose_output_slave(out_bundle->bond, &ctx->flow,
6949 /* No slaves enabled, so drop packet. */
6954 old_tci = ctx->flow.vlan_tci;
6956 if (tci || out_bundle->use_priority_tags) {
6957 tci |= ctx->flow.vlan_tci & htons(VLAN_PCP_MASK);
6959 tci |= htons(VLAN_CFI);
6962 ctx->flow.vlan_tci = tci;
6964 compose_output_action(ctx, port->up.ofp_port);
6965 ctx->flow.vlan_tci = old_tci;
6969 mirror_mask_ffs(mirror_mask_t mask)
6971 BUILD_ASSERT_DECL(sizeof(unsigned int) >= sizeof(mask));
6976 ofbundle_trunks_vlan(const struct ofbundle *bundle, uint16_t vlan)
6978 return (bundle->vlan_mode != PORT_VLAN_ACCESS
6979 && (!bundle->trunks || bitmap_is_set(bundle->trunks, vlan)));
6983 ofbundle_includes_vlan(const struct ofbundle *bundle, uint16_t vlan)
6985 return vlan == bundle->vlan || ofbundle_trunks_vlan(bundle, vlan);
6988 /* Returns an arbitrary interface within 'bundle'. */
6989 static struct ofport_dpif *
6990 ofbundle_get_a_port(const struct ofbundle *bundle)
6992 return CONTAINER_OF(list_front(&bundle->ports),
6993 struct ofport_dpif, bundle_node);
6997 vlan_is_mirrored(const struct ofmirror *m, int vlan)
6999 return !m->vlans || bitmap_is_set(m->vlans, vlan);
7003 add_mirror_actions(struct action_xlate_ctx *ctx, const struct flow *orig_flow)
7005 struct ofproto_dpif *ofproto = ctx->ofproto;
7006 mirror_mask_t mirrors;
7007 struct ofbundle *in_bundle;
7010 const struct nlattr *a;
7013 in_bundle = lookup_input_bundle(ctx->ofproto, orig_flow->in_port,
7014 ctx->packet != NULL, NULL);
7018 mirrors = in_bundle->src_mirrors;
7020 /* Drop frames on bundles reserved for mirroring. */
7021 if (in_bundle->mirror_out) {
7022 if (ctx->packet != NULL) {
7023 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7024 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7025 "%s, which is reserved exclusively for mirroring",
7026 ctx->ofproto->up.name, in_bundle->name);
7032 vid = vlan_tci_to_vid(orig_flow->vlan_tci);
7033 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
7036 vlan = input_vid_to_vlan(in_bundle, vid);
7038 /* Look at the output ports to check for destination selections. */
7040 NL_ATTR_FOR_EACH (a, left, ctx->odp_actions->data,
7041 ctx->odp_actions->size) {
7042 enum ovs_action_attr type = nl_attr_type(a);
7043 struct ofport_dpif *ofport;
7045 if (type != OVS_ACTION_ATTR_OUTPUT) {
7049 ofport = get_odp_port(ofproto, nl_attr_get_u32(a));
7050 if (ofport && ofport->bundle) {
7051 mirrors |= ofport->bundle->dst_mirrors;
7059 /* Restore the original packet before adding the mirror actions. */
7060 ctx->flow = *orig_flow;
7065 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7067 if (!vlan_is_mirrored(m, vlan)) {
7068 mirrors = zero_rightmost_1bit(mirrors);
7072 mirrors &= ~m->dup_mirrors;
7073 ctx->mirrors |= m->dup_mirrors;
7075 output_normal(ctx, m->out, vlan);
7076 } else if (vlan != m->out_vlan
7077 && !eth_addr_is_reserved(orig_flow->dl_dst)) {
7078 struct ofbundle *bundle;
7080 HMAP_FOR_EACH (bundle, hmap_node, &ofproto->bundles) {
7081 if (ofbundle_includes_vlan(bundle, m->out_vlan)
7082 && !bundle->mirror_out) {
7083 output_normal(ctx, bundle, m->out_vlan);
7091 update_mirror_stats(struct ofproto_dpif *ofproto, mirror_mask_t mirrors,
7092 uint64_t packets, uint64_t bytes)
7098 for (; mirrors; mirrors = zero_rightmost_1bit(mirrors)) {
7101 m = ofproto->mirrors[mirror_mask_ffs(mirrors) - 1];
7104 /* In normal circumstances 'm' will not be NULL. However,
7105 * if mirrors are reconfigured, we can temporarily get out
7106 * of sync in facet_revalidate(). We could "correct" the
7107 * mirror list before reaching here, but doing that would
7108 * not properly account the traffic stats we've currently
7109 * accumulated for previous mirror configuration. */
7113 m->packet_count += packets;
7114 m->byte_count += bytes;
7118 /* A VM broadcasts a gratuitous ARP to indicate that it has resumed after
7119 * migration. Older Citrix-patched Linux DomU used gratuitous ARP replies to
7120 * indicate this; newer upstream kernels use gratuitous ARP requests. */
7122 is_gratuitous_arp(const struct flow *flow)
7124 return (flow->dl_type == htons(ETH_TYPE_ARP)
7125 && eth_addr_is_broadcast(flow->dl_dst)
7126 && (flow->nw_proto == ARP_OP_REPLY
7127 || (flow->nw_proto == ARP_OP_REQUEST
7128 && flow->nw_src == flow->nw_dst)));
7132 update_learning_table(struct ofproto_dpif *ofproto,
7133 const struct flow *flow, int vlan,
7134 struct ofbundle *in_bundle)
7136 struct mac_entry *mac;
7138 /* Don't learn the OFPP_NONE port. */
7139 if (in_bundle == &ofpp_none_bundle) {
7143 if (!mac_learning_may_learn(ofproto->ml, flow->dl_src, vlan)) {
7147 mac = mac_learning_insert(ofproto->ml, flow->dl_src, vlan);
7148 if (is_gratuitous_arp(flow)) {
7149 /* We don't want to learn from gratuitous ARP packets that are
7150 * reflected back over bond slaves so we lock the learning table. */
7151 if (!in_bundle->bond) {
7152 mac_entry_set_grat_arp_lock(mac);
7153 } else if (mac_entry_is_grat_arp_locked(mac)) {
7158 if (mac_entry_is_new(mac) || mac->port.p != in_bundle) {
7159 /* The log messages here could actually be useful in debugging,
7160 * so keep the rate limit relatively high. */
7161 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(30, 300);
7162 VLOG_DBG_RL(&rl, "bridge %s: learned that "ETH_ADDR_FMT" is "
7163 "on port %s in VLAN %d",
7164 ofproto->up.name, ETH_ADDR_ARGS(flow->dl_src),
7165 in_bundle->name, vlan);
7167 mac->port.p = in_bundle;
7168 tag_set_add(&ofproto->backer->revalidate_set,
7169 mac_learning_changed(ofproto->ml, mac));
7173 static struct ofbundle *
7174 lookup_input_bundle(const struct ofproto_dpif *ofproto, uint16_t in_port,
7175 bool warn, struct ofport_dpif **in_ofportp)
7177 struct ofport_dpif *ofport;
7179 /* Find the port and bundle for the received packet. */
7180 ofport = get_ofp_port(ofproto, in_port);
7182 *in_ofportp = ofport;
7184 if (ofport && ofport->bundle) {
7185 return ofport->bundle;
7188 /* Special-case OFPP_NONE, which a controller may use as the ingress
7189 * port for traffic that it is sourcing. */
7190 if (in_port == OFPP_NONE) {
7191 return &ofpp_none_bundle;
7194 /* Odd. A few possible reasons here:
7196 * - We deleted a port but there are still a few packets queued up
7199 * - Someone externally added a port (e.g. "ovs-dpctl add-if") that
7200 * we don't know about.
7202 * - The ofproto client didn't configure the port as part of a bundle.
7203 * This is particularly likely to happen if a packet was received on the
7204 * port after it was created, but before the client had a chance to
7205 * configure its bundle.
7208 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7210 VLOG_WARN_RL(&rl, "bridge %s: received packet on unknown "
7211 "port %"PRIu16, ofproto->up.name, in_port);
7216 /* Determines whether packets in 'flow' within 'ofproto' should be forwarded or
7217 * dropped. Returns true if they may be forwarded, false if they should be
7220 * 'in_port' must be the ofport_dpif that corresponds to flow->in_port.
7221 * 'in_port' must be part of a bundle (e.g. in_port->bundle must be nonnull).
7223 * 'vlan' must be the VLAN that corresponds to flow->vlan_tci on 'in_port', as
7224 * returned by input_vid_to_vlan(). It must be a valid VLAN for 'in_port', as
7225 * checked by input_vid_is_valid().
7227 * May also add tags to '*tags', although the current implementation only does
7228 * so in one special case.
7231 is_admissible(struct action_xlate_ctx *ctx, struct ofport_dpif *in_port,
7234 struct ofproto_dpif *ofproto = ctx->ofproto;
7235 struct flow *flow = &ctx->flow;
7236 struct ofbundle *in_bundle = in_port->bundle;
7238 /* Drop frames for reserved multicast addresses
7239 * only if forward_bpdu option is absent. */
7240 if (!ofproto->up.forward_bpdu && eth_addr_is_reserved(flow->dl_dst)) {
7241 xlate_report(ctx, "packet has reserved destination MAC, dropping");
7245 if (in_bundle->bond) {
7246 struct mac_entry *mac;
7248 switch (bond_check_admissibility(in_bundle->bond, in_port,
7249 flow->dl_dst, &ctx->tags)) {
7254 xlate_report(ctx, "bonding refused admissibility, dropping");
7257 case BV_DROP_IF_MOVED:
7258 mac = mac_learning_lookup(ofproto->ml, flow->dl_src, vlan, NULL);
7259 if (mac && mac->port.p != in_bundle &&
7260 (!is_gratuitous_arp(flow)
7261 || mac_entry_is_grat_arp_locked(mac))) {
7262 xlate_report(ctx, "SLB bond thinks this packet looped back, "
7274 xlate_normal(struct action_xlate_ctx *ctx)
7276 struct ofport_dpif *in_port;
7277 struct ofbundle *in_bundle;
7278 struct mac_entry *mac;
7282 ctx->has_normal = true;
7284 in_bundle = lookup_input_bundle(ctx->ofproto, ctx->flow.in_port,
7285 ctx->packet != NULL, &in_port);
7287 xlate_report(ctx, "no input bundle, dropping");
7291 /* Drop malformed frames. */
7292 if (ctx->flow.dl_type == htons(ETH_TYPE_VLAN) &&
7293 !(ctx->flow.vlan_tci & htons(VLAN_CFI))) {
7294 if (ctx->packet != NULL) {
7295 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7296 VLOG_WARN_RL(&rl, "bridge %s: dropping packet with partial "
7297 "VLAN tag received on port %s",
7298 ctx->ofproto->up.name, in_bundle->name);
7300 xlate_report(ctx, "partial VLAN tag, dropping");
7304 /* Drop frames on bundles reserved for mirroring. */
7305 if (in_bundle->mirror_out) {
7306 if (ctx->packet != NULL) {
7307 static struct vlog_rate_limit rl = VLOG_RATE_LIMIT_INIT(1, 5);
7308 VLOG_WARN_RL(&rl, "bridge %s: dropping packet received on port "
7309 "%s, which is reserved exclusively for mirroring",
7310 ctx->ofproto->up.name, in_bundle->name);
7312 xlate_report(ctx, "input port is mirror output port, dropping");
7317 vid = vlan_tci_to_vid(ctx->flow.vlan_tci);
7318 if (!input_vid_is_valid(vid, in_bundle, ctx->packet != NULL)) {
7319 xlate_report(ctx, "disallowed VLAN VID for this input port, dropping");
7322 vlan = input_vid_to_vlan(in_bundle, vid);
7324 /* Check other admissibility requirements. */
7325 if (in_port && !is_admissible(ctx, in_port, vlan)) {
7329 /* Learn source MAC. */
7330 if (ctx->may_learn) {
7331 update_learning_table(ctx->ofproto, &ctx->flow, vlan, in_bundle);
7334 /* Determine output bundle. */
7335 mac = mac_learning_lookup(ctx->ofproto->ml, ctx->flow.dl_dst, vlan,
7338 if (mac->port.p != in_bundle) {
7339 xlate_report(ctx, "forwarding to learned port");
7340 output_normal(ctx, mac->port.p, vlan);
7342 xlate_report(ctx, "learned port is input port, dropping");
7345 struct ofbundle *bundle;
7347 xlate_report(ctx, "no learned MAC for destination, flooding");
7348 HMAP_FOR_EACH (bundle, hmap_node, &ctx->ofproto->bundles) {
7349 if (bundle != in_bundle
7350 && ofbundle_includes_vlan(bundle, vlan)
7351 && bundle->floodable
7352 && !bundle->mirror_out) {
7353 output_normal(ctx, bundle, vlan);
7356 ctx->nf_output_iface = NF_OUT_FLOOD;
7360 /* Optimized flow revalidation.
7362 * It's a difficult problem, in general, to tell which facets need to have
7363 * their actions recalculated whenever the OpenFlow flow table changes. We
7364 * don't try to solve that general problem: for most kinds of OpenFlow flow
7365 * table changes, we recalculate the actions for every facet. This is
7366 * relatively expensive, but it's good enough if the OpenFlow flow table
7367 * doesn't change very often.
7369 * However, we can expect one particular kind of OpenFlow flow table change to
7370 * happen frequently: changes caused by MAC learning. To avoid wasting a lot
7371 * of CPU on revalidating every facet whenever MAC learning modifies the flow
7372 * table, we add a special case that applies to flow tables in which every rule
7373 * has the same form (that is, the same wildcards), except that the table is
7374 * also allowed to have a single "catch-all" flow that matches all packets. We
7375 * optimize this case by tagging all of the facets that resubmit into the table
7376 * and invalidating the same tag whenever a flow changes in that table. The
7377 * end result is that we revalidate just the facets that need it (and sometimes
7378 * a few more, but not all of the facets or even all of the facets that
7379 * resubmit to the table modified by MAC learning). */
7381 /* Calculates the tag to use for 'flow' and mask 'mask' when it is inserted
7382 * into an OpenFlow table with the given 'basis'. */
7384 rule_calculate_tag(const struct flow *flow, const struct minimask *mask,
7387 if (minimask_is_catchall(mask)) {
7390 uint32_t hash = flow_hash_in_minimask(flow, mask, secret);
7391 return tag_create_deterministic(hash);
7395 /* Following a change to OpenFlow table 'table_id' in 'ofproto', update the
7396 * taggability of that table.
7398 * This function must be called after *each* change to a flow table. If you
7399 * skip calling it on some changes then the pointer comparisons at the end can
7400 * be invalid if you get unlucky. For example, if a flow removal causes a
7401 * cls_table to be destroyed and then a flow insertion causes a cls_table with
7402 * different wildcards to be created with the same address, then this function
7403 * will incorrectly skip revalidation. */
7405 table_update_taggable(struct ofproto_dpif *ofproto, uint8_t table_id)
7407 struct table_dpif *table = &ofproto->tables[table_id];
7408 const struct oftable *oftable = &ofproto->up.tables[table_id];
7409 struct cls_table *catchall, *other;
7410 struct cls_table *t;
7412 catchall = other = NULL;
7414 switch (hmap_count(&oftable->cls.tables)) {
7416 /* We could tag this OpenFlow table but it would make the logic a
7417 * little harder and it's a corner case that doesn't seem worth it
7423 HMAP_FOR_EACH (t, hmap_node, &oftable->cls.tables) {
7424 if (cls_table_is_catchall(t)) {
7426 } else if (!other) {
7429 /* Indicate that we can't tag this by setting both tables to
7430 * NULL. (We know that 'catchall' is already NULL.) */
7437 /* Can't tag this table. */
7441 if (table->catchall_table != catchall || table->other_table != other) {
7442 table->catchall_table = catchall;
7443 table->other_table = other;
7444 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7448 /* Given 'rule' that has changed in some way (either it is a rule being
7449 * inserted, a rule being deleted, or a rule whose actions are being
7450 * modified), marks facets for revalidation to ensure that packets will be
7451 * forwarded correctly according to the new state of the flow table.
7453 * This function must be called after *each* change to a flow table. See
7454 * the comment on table_update_taggable() for more information. */
7456 rule_invalidate(const struct rule_dpif *rule)
7458 struct ofproto_dpif *ofproto = ofproto_dpif_cast(rule->up.ofproto);
7460 table_update_taggable(ofproto, rule->up.table_id);
7462 if (!ofproto->backer->need_revalidate) {
7463 struct table_dpif *table = &ofproto->tables[rule->up.table_id];
7465 if (table->other_table && rule->tag) {
7466 tag_set_add(&ofproto->backer->revalidate_set, rule->tag);
7468 ofproto->backer->need_revalidate = REV_FLOW_TABLE;
7474 set_frag_handling(struct ofproto *ofproto_,
7475 enum ofp_config_flags frag_handling)
7477 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7478 if (frag_handling != OFPC_FRAG_REASM) {
7479 ofproto->backer->need_revalidate = REV_RECONFIGURE;
7487 packet_out(struct ofproto *ofproto_, struct ofpbuf *packet,
7488 const struct flow *flow,
7489 const struct ofpact *ofpacts, size_t ofpacts_len)
7491 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7492 struct initial_vals initial_vals;
7493 struct odputil_keybuf keybuf;
7494 struct dpif_flow_stats stats;
7498 struct action_xlate_ctx ctx;
7499 uint64_t odp_actions_stub[1024 / 8];
7500 struct ofpbuf odp_actions;
7502 ofpbuf_use_stack(&key, &keybuf, sizeof keybuf);
7503 odp_flow_key_from_flow(&key, flow,
7504 ofp_port_to_odp_port(ofproto, flow->in_port));
7506 dpif_flow_stats_extract(flow, packet, time_msec(), &stats);
7508 initial_vals.vlan_tci = flow->vlan_tci;
7509 initial_vals.tunnel_ip_tos = 0;
7510 action_xlate_ctx_init(&ctx, ofproto, flow, &initial_vals, NULL,
7511 packet_get_tcp_flags(packet, flow), packet);
7512 ctx.resubmit_stats = &stats;
7514 ofpbuf_use_stub(&odp_actions,
7515 odp_actions_stub, sizeof odp_actions_stub);
7516 xlate_actions(&ctx, ofpacts, ofpacts_len, &odp_actions);
7517 dpif_execute(ofproto->backer->dpif, key.data, key.size,
7518 odp_actions.data, odp_actions.size, packet);
7519 ofpbuf_uninit(&odp_actions);
7527 set_netflow(struct ofproto *ofproto_,
7528 const struct netflow_options *netflow_options)
7530 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7532 if (netflow_options) {
7533 if (!ofproto->netflow) {
7534 ofproto->netflow = netflow_create();
7536 return netflow_set_options(ofproto->netflow, netflow_options);
7538 netflow_destroy(ofproto->netflow);
7539 ofproto->netflow = NULL;
7545 get_netflow_ids(const struct ofproto *ofproto_,
7546 uint8_t *engine_type, uint8_t *engine_id)
7548 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofproto_);
7550 dpif_get_netflow_ids(ofproto->backer->dpif, engine_type, engine_id);
7554 send_active_timeout(struct ofproto_dpif *ofproto, struct facet *facet)
7556 if (!facet_is_controller_flow(facet) &&
7557 netflow_active_timeout_expired(ofproto->netflow, &facet->nf_flow)) {
7558 struct subfacet *subfacet;
7559 struct ofexpired expired;
7561 LIST_FOR_EACH (subfacet, list_node, &facet->subfacets) {
7562 if (subfacet->path == SF_FAST_PATH) {
7563 struct dpif_flow_stats stats;
7565 subfacet_reinstall(subfacet, &stats);
7566 subfacet_update_stats(subfacet, &stats);
7570 expired.flow = facet->flow;
7571 expired.packet_count = facet->packet_count;
7572 expired.byte_count = facet->byte_count;
7573 expired.used = facet->used;
7574 netflow_expire(ofproto->netflow, &facet->nf_flow, &expired);
7579 send_netflow_active_timeouts(struct ofproto_dpif *ofproto)
7581 struct facet *facet;
7583 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7584 send_active_timeout(ofproto, facet);
7588 static struct ofproto_dpif *
7589 ofproto_dpif_lookup(const char *name)
7591 struct ofproto_dpif *ofproto;
7593 HMAP_FOR_EACH_WITH_HASH (ofproto, all_ofproto_dpifs_node,
7594 hash_string(name, 0), &all_ofproto_dpifs) {
7595 if (!strcmp(ofproto->up.name, name)) {
7603 ofproto_unixctl_fdb_flush(struct unixctl_conn *conn, int argc,
7604 const char *argv[], void *aux OVS_UNUSED)
7606 struct ofproto_dpif *ofproto;
7609 ofproto = ofproto_dpif_lookup(argv[1]);
7611 unixctl_command_reply_error(conn, "no such bridge");
7614 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7616 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
7617 mac_learning_flush(ofproto->ml, &ofproto->backer->revalidate_set);
7621 unixctl_command_reply(conn, "table successfully flushed");
7625 ofproto_unixctl_fdb_show(struct unixctl_conn *conn, int argc OVS_UNUSED,
7626 const char *argv[], void *aux OVS_UNUSED)
7628 struct ds ds = DS_EMPTY_INITIALIZER;
7629 const struct ofproto_dpif *ofproto;
7630 const struct mac_entry *e;
7632 ofproto = ofproto_dpif_lookup(argv[1]);
7634 unixctl_command_reply_error(conn, "no such bridge");
7638 ds_put_cstr(&ds, " port VLAN MAC Age\n");
7639 LIST_FOR_EACH (e, lru_node, &ofproto->ml->lrus) {
7640 struct ofbundle *bundle = e->port.p;
7641 ds_put_format(&ds, "%5d %4d "ETH_ADDR_FMT" %3d\n",
7642 ofbundle_get_a_port(bundle)->odp_port,
7643 e->vlan, ETH_ADDR_ARGS(e->mac),
7644 mac_entry_age(ofproto->ml, e));
7646 unixctl_command_reply(conn, ds_cstr(&ds));
7651 struct action_xlate_ctx ctx;
7657 trace_format_rule(struct ds *result, uint8_t table_id, int level,
7658 const struct rule_dpif *rule)
7660 ds_put_char_multiple(result, '\t', level);
7662 ds_put_cstr(result, "No match\n");
7666 ds_put_format(result, "Rule: table=%"PRIu8" cookie=%#"PRIx64" ",
7667 table_id, ntohll(rule->up.flow_cookie));
7668 cls_rule_format(&rule->up.cr, result);
7669 ds_put_char(result, '\n');
7671 ds_put_char_multiple(result, '\t', level);
7672 ds_put_cstr(result, "OpenFlow ");
7673 ofpacts_format(rule->up.ofpacts, rule->up.ofpacts_len, result);
7674 ds_put_char(result, '\n');
7678 trace_format_flow(struct ds *result, int level, const char *title,
7679 struct trace_ctx *trace)
7681 ds_put_char_multiple(result, '\t', level);
7682 ds_put_format(result, "%s: ", title);
7683 if (flow_equal(&trace->ctx.flow, &trace->flow)) {
7684 ds_put_cstr(result, "unchanged");
7686 flow_format(result, &trace->ctx.flow);
7687 trace->flow = trace->ctx.flow;
7689 ds_put_char(result, '\n');
7693 trace_format_regs(struct ds *result, int level, const char *title,
7694 struct trace_ctx *trace)
7698 ds_put_char_multiple(result, '\t', level);
7699 ds_put_format(result, "%s:", title);
7700 for (i = 0; i < FLOW_N_REGS; i++) {
7701 ds_put_format(result, " reg%zu=0x%"PRIx32, i, trace->flow.regs[i]);
7703 ds_put_char(result, '\n');
7707 trace_format_odp(struct ds *result, int level, const char *title,
7708 struct trace_ctx *trace)
7710 struct ofpbuf *odp_actions = trace->ctx.odp_actions;
7712 ds_put_char_multiple(result, '\t', level);
7713 ds_put_format(result, "%s: ", title);
7714 format_odp_actions(result, odp_actions->data, odp_actions->size);
7715 ds_put_char(result, '\n');
7719 trace_resubmit(struct action_xlate_ctx *ctx, struct rule_dpif *rule)
7721 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7722 struct ds *result = trace->result;
7724 ds_put_char(result, '\n');
7725 trace_format_flow(result, ctx->recurse + 1, "Resubmitted flow", trace);
7726 trace_format_regs(result, ctx->recurse + 1, "Resubmitted regs", trace);
7727 trace_format_odp(result, ctx->recurse + 1, "Resubmitted odp", trace);
7728 trace_format_rule(result, ctx->table_id, ctx->recurse + 1, rule);
7732 trace_report(struct action_xlate_ctx *ctx, const char *s)
7734 struct trace_ctx *trace = CONTAINER_OF(ctx, struct trace_ctx, ctx);
7735 struct ds *result = trace->result;
7737 ds_put_char_multiple(result, '\t', ctx->recurse);
7738 ds_put_cstr(result, s);
7739 ds_put_char(result, '\n');
7743 ofproto_unixctl_trace(struct unixctl_conn *conn, int argc, const char *argv[],
7744 void *aux OVS_UNUSED)
7746 const char *dpname = argv[1];
7747 struct ofproto_dpif *ofproto;
7748 struct ofpbuf odp_key;
7749 struct ofpbuf *packet;
7750 struct initial_vals initial_vals;
7756 ofpbuf_init(&odp_key, 0);
7759 ofproto = ofproto_dpif_lookup(dpname);
7761 unixctl_command_reply_error(conn, "Unknown ofproto (use ofproto/list "
7765 if (argc == 3 || (argc == 4 && !strcmp(argv[3], "-generate"))) {
7766 /* ofproto/trace dpname flow [-generate] */
7767 const char *flow_s = argv[2];
7768 const char *generate_s = argv[3];
7770 /* Allow 'flow_s' to be either a datapath flow or an OpenFlow-like
7771 * flow. We guess which type it is based on whether 'flow_s' contains
7772 * an '(', since a datapath flow always contains '(') but an
7773 * OpenFlow-like flow should not (in fact it's allowed but I believe
7774 * that's not documented anywhere).
7776 * An alternative would be to try to parse 'flow_s' both ways, but then
7777 * it would be tricky giving a sensible error message. After all, do
7778 * you just say "syntax error" or do you present both error messages?
7779 * Both choices seem lousy. */
7780 if (strchr(flow_s, '(')) {
7783 /* Convert string to datapath key. */
7784 ofpbuf_init(&odp_key, 0);
7785 error = odp_flow_key_from_string(flow_s, NULL, &odp_key);
7787 unixctl_command_reply_error(conn, "Bad flow syntax");
7791 /* The user might have specified the wrong ofproto but within the
7792 * same backer. That's OK, ofproto_receive() can find the right
7794 if (ofproto_receive(ofproto->backer, NULL, odp_key.data,
7795 odp_key.size, &flow, NULL, &ofproto, NULL,
7797 unixctl_command_reply_error(conn, "Invalid flow");
7800 ds_put_format(&result, "Bridge: %s\n", ofproto->up.name);
7804 error_s = parse_ofp_exact_flow(&flow, argv[2]);
7806 unixctl_command_reply_error(conn, error_s);
7811 initial_vals.vlan_tci = flow.vlan_tci;
7812 initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos;
7815 /* Generate a packet, if requested. */
7817 packet = ofpbuf_new(0);
7818 flow_compose(packet, &flow);
7820 } else if (argc == 7) {
7821 /* ofproto/trace dpname priority tun_id in_port mark packet */
7822 const char *priority_s = argv[2];
7823 const char *tun_id_s = argv[3];
7824 const char *in_port_s = argv[4];
7825 const char *mark_s = argv[5];
7826 const char *packet_s = argv[6];
7827 uint32_t in_port = atoi(in_port_s);
7828 ovs_be64 tun_id = htonll(strtoull(tun_id_s, NULL, 0));
7829 uint32_t priority = atoi(priority_s);
7830 uint32_t mark = atoi(mark_s);
7833 msg = eth_from_hex(packet_s, &packet);
7835 unixctl_command_reply_error(conn, msg);
7839 ds_put_cstr(&result, "Packet: ");
7840 s = ofp_packet_to_string(packet->data, packet->size);
7841 ds_put_cstr(&result, s);
7844 flow_extract(packet, priority, mark, NULL, in_port, &flow);
7845 flow.tunnel.tun_id = tun_id;
7846 initial_vals.vlan_tci = flow.vlan_tci;
7847 initial_vals.tunnel_ip_tos = flow.tunnel.ip_tos;
7849 unixctl_command_reply_error(conn, "Bad command syntax");
7853 ofproto_trace(ofproto, &flow, packet, &initial_vals, &result);
7854 unixctl_command_reply(conn, ds_cstr(&result));
7857 ds_destroy(&result);
7858 ofpbuf_delete(packet);
7859 ofpbuf_uninit(&odp_key);
7863 ofproto_trace(struct ofproto_dpif *ofproto, const struct flow *flow,
7864 const struct ofpbuf *packet,
7865 const struct initial_vals *initial_vals, struct ds *ds)
7867 struct rule_dpif *rule;
7869 ds_put_cstr(ds, "Flow: ");
7870 flow_format(ds, flow);
7871 ds_put_char(ds, '\n');
7873 rule = rule_dpif_lookup(ofproto, flow);
7875 trace_format_rule(ds, 0, 0, rule);
7876 if (rule == ofproto->miss_rule) {
7877 ds_put_cstr(ds, "\nNo match, flow generates \"packet in\"s.\n");
7878 } else if (rule == ofproto->no_packet_in_rule) {
7879 ds_put_cstr(ds, "\nNo match, packets dropped because "
7880 "OFPPC_NO_PACKET_IN is set on in_port.\n");
7884 uint64_t odp_actions_stub[1024 / 8];
7885 struct ofpbuf odp_actions;
7887 struct trace_ctx trace;
7890 tcp_flags = packet ? packet_get_tcp_flags(packet, flow) : 0;
7893 ofpbuf_use_stub(&odp_actions,
7894 odp_actions_stub, sizeof odp_actions_stub);
7895 action_xlate_ctx_init(&trace.ctx, ofproto, flow, initial_vals,
7896 rule, tcp_flags, packet);
7897 trace.ctx.resubmit_hook = trace_resubmit;
7898 trace.ctx.report_hook = trace_report;
7899 xlate_actions(&trace.ctx, rule->up.ofpacts, rule->up.ofpacts_len,
7902 ds_put_char(ds, '\n');
7903 trace_format_flow(ds, 0, "Final flow", &trace);
7904 ds_put_cstr(ds, "Datapath actions: ");
7905 format_odp_actions(ds, odp_actions.data, odp_actions.size);
7906 ofpbuf_uninit(&odp_actions);
7908 if (trace.ctx.slow) {
7909 enum slow_path_reason slow;
7911 ds_put_cstr(ds, "\nThis flow is handled by the userspace "
7912 "slow path because it:");
7913 for (slow = trace.ctx.slow; slow; ) {
7914 enum slow_path_reason bit = rightmost_1bit(slow);
7918 ds_put_cstr(ds, "\n\t- Consists of CFM packets.");
7921 ds_put_cstr(ds, "\n\t- Consists of LACP packets.");
7924 ds_put_cstr(ds, "\n\t- Consists of STP packets.");
7927 ds_put_cstr(ds, "\n\t- Needs in-band special case "
7930 ds_put_cstr(ds, "\n\t (The datapath actions are "
7931 "incomplete--for complete actions, "
7932 "please supply a packet.)");
7935 case SLOW_CONTROLLER:
7936 ds_put_cstr(ds, "\n\t- Sends \"packet-in\" messages "
7937 "to the OpenFlow controller.");
7940 ds_put_cstr(ds, "\n\t- Needs more specific matching "
7941 "than the datapath supports.");
7948 if (slow & ~SLOW_MATCH) {
7949 ds_put_cstr(ds, "\nThe datapath actions above do not reflect "
7950 "the special slow-path processing.");
7957 ofproto_dpif_clog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7958 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7961 unixctl_command_reply(conn, NULL);
7965 ofproto_dpif_unclog(struct unixctl_conn *conn OVS_UNUSED, int argc OVS_UNUSED,
7966 const char *argv[] OVS_UNUSED, void *aux OVS_UNUSED)
7969 unixctl_command_reply(conn, NULL);
7972 /* Runs a self-check of flow translations in 'ofproto'. Appends a message to
7973 * 'reply' describing the results. */
7975 ofproto_dpif_self_check__(struct ofproto_dpif *ofproto, struct ds *reply)
7977 struct facet *facet;
7981 HMAP_FOR_EACH (facet, hmap_node, &ofproto->facets) {
7982 if (!facet_check_consistency(facet)) {
7987 ofproto->backer->need_revalidate = REV_INCONSISTENCY;
7991 ds_put_format(reply, "%s: self-check failed (%d errors)\n",
7992 ofproto->up.name, errors);
7994 ds_put_format(reply, "%s: self-check passed\n", ofproto->up.name);
7999 ofproto_dpif_self_check(struct unixctl_conn *conn,
8000 int argc, const char *argv[], void *aux OVS_UNUSED)
8002 struct ds reply = DS_EMPTY_INITIALIZER;
8003 struct ofproto_dpif *ofproto;
8006 ofproto = ofproto_dpif_lookup(argv[1]);
8008 unixctl_command_reply_error(conn, "Unknown ofproto (use "
8009 "ofproto/list for help)");
8012 ofproto_dpif_self_check__(ofproto, &reply);
8014 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8015 ofproto_dpif_self_check__(ofproto, &reply);
8019 unixctl_command_reply(conn, ds_cstr(&reply));
8023 /* Store the current ofprotos in 'ofproto_shash'. Returns a sorted list
8024 * of the 'ofproto_shash' nodes. It is the responsibility of the caller
8025 * to destroy 'ofproto_shash' and free the returned value. */
8026 static const struct shash_node **
8027 get_ofprotos(struct shash *ofproto_shash)
8029 const struct ofproto_dpif *ofproto;
8031 HMAP_FOR_EACH (ofproto, all_ofproto_dpifs_node, &all_ofproto_dpifs) {
8032 char *name = xasprintf("%s@%s", ofproto->up.type, ofproto->up.name);
8033 shash_add_nocopy(ofproto_shash, name, ofproto);
8036 return shash_sort(ofproto_shash);
8040 ofproto_unixctl_dpif_dump_dps(struct unixctl_conn *conn, int argc OVS_UNUSED,
8041 const char *argv[] OVS_UNUSED,
8042 void *aux OVS_UNUSED)
8044 struct ds ds = DS_EMPTY_INITIALIZER;
8045 struct shash ofproto_shash;
8046 const struct shash_node **sorted_ofprotos;
8049 shash_init(&ofproto_shash);
8050 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8051 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8052 const struct shash_node *node = sorted_ofprotos[i];
8053 ds_put_format(&ds, "%s\n", node->name);
8056 shash_destroy(&ofproto_shash);
8057 free(sorted_ofprotos);
8059 unixctl_command_reply(conn, ds_cstr(&ds));
8064 show_dp_format(const struct ofproto_dpif *ofproto, struct ds *ds)
8066 const struct shash_node **ports;
8069 ds_put_format(ds, "%s (%s):\n", ofproto->up.name,
8070 dpif_name(ofproto->backer->dpif));
8072 "\tlookups: hit:%"PRIu64" missed:%"PRIu64"\n",
8073 ofproto->n_hit, ofproto->n_missed);
8074 ds_put_format(ds, "\tflows: %zu\n",
8075 hmap_count(&ofproto->subfacets));
8077 ports = shash_sort(&ofproto->up.port_by_name);
8078 for (i = 0; i < shash_count(&ofproto->up.port_by_name); i++) {
8079 const struct shash_node *node = ports[i];
8080 struct ofport *ofport = node->data;
8081 const char *name = netdev_get_name(ofport->netdev);
8082 const char *type = netdev_get_type(ofport->netdev);
8085 ds_put_format(ds, "\t%s %u/", name, ofport->ofp_port);
8087 odp_port = ofp_port_to_odp_port(ofproto, ofport->ofp_port);
8088 if (odp_port != OVSP_NONE) {
8089 ds_put_format(ds, "%"PRIu32":", odp_port);
8091 ds_put_cstr(ds, "none:");
8094 if (strcmp(type, "system")) {
8095 struct netdev *netdev;
8098 ds_put_format(ds, " (%s", type);
8100 error = netdev_open(name, type, &netdev);
8105 error = netdev_get_config(netdev, &config);
8107 const struct smap_node **nodes;
8110 nodes = smap_sort(&config);
8111 for (i = 0; i < smap_count(&config); i++) {
8112 const struct smap_node *node = nodes[i];
8113 ds_put_format(ds, "%c %s=%s", i ? ',' : ':',
8114 node->key, node->value);
8118 smap_destroy(&config);
8120 netdev_close(netdev);
8122 ds_put_char(ds, ')');
8124 ds_put_char(ds, '\n');
8130 ofproto_unixctl_dpif_show(struct unixctl_conn *conn, int argc,
8131 const char *argv[], void *aux OVS_UNUSED)
8133 struct ds ds = DS_EMPTY_INITIALIZER;
8134 const struct ofproto_dpif *ofproto;
8138 for (i = 1; i < argc; i++) {
8139 ofproto = ofproto_dpif_lookup(argv[i]);
8141 ds_put_format(&ds, "Unknown bridge %s (use dpif/dump-dps "
8142 "for help)", argv[i]);
8143 unixctl_command_reply_error(conn, ds_cstr(&ds));
8146 show_dp_format(ofproto, &ds);
8149 struct shash ofproto_shash;
8150 const struct shash_node **sorted_ofprotos;
8153 shash_init(&ofproto_shash);
8154 sorted_ofprotos = get_ofprotos(&ofproto_shash);
8155 for (i = 0; i < shash_count(&ofproto_shash); i++) {
8156 const struct shash_node *node = sorted_ofprotos[i];
8157 show_dp_format(node->data, &ds);
8160 shash_destroy(&ofproto_shash);
8161 free(sorted_ofprotos);
8164 unixctl_command_reply(conn, ds_cstr(&ds));
8169 ofproto_unixctl_dpif_dump_flows(struct unixctl_conn *conn,
8170 int argc OVS_UNUSED, const char *argv[],
8171 void *aux OVS_UNUSED)
8173 struct ds ds = DS_EMPTY_INITIALIZER;
8174 const struct ofproto_dpif *ofproto;
8175 struct subfacet *subfacet;
8177 ofproto = ofproto_dpif_lookup(argv[1]);
8179 unixctl_command_reply_error(conn, "no such bridge");
8183 update_stats(ofproto->backer);
8185 HMAP_FOR_EACH (subfacet, hmap_node, &ofproto->subfacets) {
8186 odp_flow_key_format(subfacet->key, subfacet->key_len, &ds);
8188 ds_put_format(&ds, ", packets:%"PRIu64", bytes:%"PRIu64", used:",
8189 subfacet->dp_packet_count, subfacet->dp_byte_count);
8190 if (subfacet->used) {
8191 ds_put_format(&ds, "%.3fs",
8192 (time_msec() - subfacet->used) / 1000.0);
8194 ds_put_format(&ds, "never");
8196 if (subfacet->facet->tcp_flags) {
8197 ds_put_cstr(&ds, ", flags:");
8198 packet_format_tcp_flags(&ds, subfacet->facet->tcp_flags);
8201 ds_put_cstr(&ds, ", actions:");
8202 if (subfacet->slow) {
8203 uint64_t slow_path_stub[128 / 8];
8204 const struct nlattr *actions;
8207 compose_slow_path(ofproto, &subfacet->facet->flow, subfacet->slow,
8208 slow_path_stub, sizeof slow_path_stub,
8209 &actions, &actions_len);
8210 format_odp_actions(&ds, actions, actions_len);
8212 format_odp_actions(&ds, subfacet->actions, subfacet->actions_len);
8214 ds_put_char(&ds, '\n');
8217 unixctl_command_reply(conn, ds_cstr(&ds));
8222 ofproto_unixctl_dpif_del_flows(struct unixctl_conn *conn,
8223 int argc OVS_UNUSED, const char *argv[],
8224 void *aux OVS_UNUSED)
8226 struct ds ds = DS_EMPTY_INITIALIZER;
8227 struct ofproto_dpif *ofproto;
8229 ofproto = ofproto_dpif_lookup(argv[1]);
8231 unixctl_command_reply_error(conn, "no such bridge");
8235 flush(&ofproto->up);
8237 unixctl_command_reply(conn, ds_cstr(&ds));
8242 ofproto_dpif_unixctl_init(void)
8244 static bool registered;
8250 unixctl_command_register(
8252 "bridge {priority tun_id in_port mark packet | odp_flow [-generate]}",
8253 2, 6, ofproto_unixctl_trace, NULL);
8254 unixctl_command_register("fdb/flush", "[bridge]", 0, 1,
8255 ofproto_unixctl_fdb_flush, NULL);
8256 unixctl_command_register("fdb/show", "bridge", 1, 1,
8257 ofproto_unixctl_fdb_show, NULL);
8258 unixctl_command_register("ofproto/clog", "", 0, 0,
8259 ofproto_dpif_clog, NULL);
8260 unixctl_command_register("ofproto/unclog", "", 0, 0,
8261 ofproto_dpif_unclog, NULL);
8262 unixctl_command_register("ofproto/self-check", "[bridge]", 0, 1,
8263 ofproto_dpif_self_check, NULL);
8264 unixctl_command_register("dpif/dump-dps", "", 0, 0,
8265 ofproto_unixctl_dpif_dump_dps, NULL);
8266 unixctl_command_register("dpif/show", "[bridge]", 0, INT_MAX,
8267 ofproto_unixctl_dpif_show, NULL);
8268 unixctl_command_register("dpif/dump-flows", "bridge", 1, 1,
8269 ofproto_unixctl_dpif_dump_flows, NULL);
8270 unixctl_command_register("dpif/del-flows", "bridge", 1, 1,
8271 ofproto_unixctl_dpif_del_flows, NULL);
8274 /* Linux VLAN device support (e.g. "eth0.10" for VLAN 10.)
8276 * This is deprecated. It is only for compatibility with broken device drivers
8277 * in old versions of Linux that do not properly support VLANs when VLAN
8278 * devices are not used. When broken device drivers are no longer in
8279 * widespread use, we will delete these interfaces. */
8282 set_realdev(struct ofport *ofport_, uint16_t realdev_ofp_port, int vid)
8284 struct ofproto_dpif *ofproto = ofproto_dpif_cast(ofport_->ofproto);
8285 struct ofport_dpif *ofport = ofport_dpif_cast(ofport_);
8287 if (realdev_ofp_port == ofport->realdev_ofp_port
8288 && vid == ofport->vlandev_vid) {
8292 ofproto->backer->need_revalidate = REV_RECONFIGURE;
8294 if (ofport->realdev_ofp_port) {
8297 if (realdev_ofp_port && ofport->bundle) {
8298 /* vlandevs are enslaved to their realdevs, so they are not allowed to
8299 * themselves be part of a bundle. */
8300 bundle_set(ofport->up.ofproto, ofport->bundle, NULL);
8303 ofport->realdev_ofp_port = realdev_ofp_port;
8304 ofport->vlandev_vid = vid;
8306 if (realdev_ofp_port) {
8307 vsp_add(ofport, realdev_ofp_port, vid);
8314 hash_realdev_vid(uint16_t realdev_ofp_port, int vid)
8316 return hash_2words(realdev_ofp_port, vid);
8319 /* Returns the ODP port number of the Linux VLAN device that corresponds to
8320 * 'vlan_tci' on the network device with port number 'realdev_odp_port' in
8321 * 'ofproto'. For example, given 'realdev_odp_port' of eth0 and 'vlan_tci' 9,
8322 * it would return the port number of eth0.9.
8324 * Unless VLAN splinters are enabled for port 'realdev_odp_port', this
8325 * function just returns its 'realdev_odp_port' argument. */
8327 vsp_realdev_to_vlandev(const struct ofproto_dpif *ofproto,
8328 uint32_t realdev_odp_port, ovs_be16 vlan_tci)
8330 if (!hmap_is_empty(&ofproto->realdev_vid_map)) {
8331 uint16_t realdev_ofp_port;
8332 int vid = vlan_tci_to_vid(vlan_tci);
8333 const struct vlan_splinter *vsp;
8335 realdev_ofp_port = odp_port_to_ofp_port(ofproto, realdev_odp_port);
8336 HMAP_FOR_EACH_WITH_HASH (vsp, realdev_vid_node,
8337 hash_realdev_vid(realdev_ofp_port, vid),
8338 &ofproto->realdev_vid_map) {
8339 if (vsp->realdev_ofp_port == realdev_ofp_port
8340 && vsp->vid == vid) {
8341 return ofp_port_to_odp_port(ofproto, vsp->vlandev_ofp_port);
8345 return realdev_odp_port;
8348 static struct vlan_splinter *
8349 vlandev_find(const struct ofproto_dpif *ofproto, uint16_t vlandev_ofp_port)
8351 struct vlan_splinter *vsp;
8353 HMAP_FOR_EACH_WITH_HASH (vsp, vlandev_node, hash_int(vlandev_ofp_port, 0),
8354 &ofproto->vlandev_map) {
8355 if (vsp->vlandev_ofp_port == vlandev_ofp_port) {
8363 /* Returns the OpenFlow port number of the "real" device underlying the Linux
8364 * VLAN device with OpenFlow port number 'vlandev_ofp_port' and stores the
8365 * VLAN VID of the Linux VLAN device in '*vid'. For example, given
8366 * 'vlandev_ofp_port' of eth0.9, it would return the OpenFlow port number of
8367 * eth0 and store 9 in '*vid'.
8369 * Returns 0 and does not modify '*vid' if 'vlandev_ofp_port' is not a Linux
8370 * VLAN device. Unless VLAN splinters are enabled, this is what this function
8373 vsp_vlandev_to_realdev(const struct ofproto_dpif *ofproto,
8374 uint16_t vlandev_ofp_port, int *vid)
8376 if (!hmap_is_empty(&ofproto->vlandev_map)) {
8377 const struct vlan_splinter *vsp;
8379 vsp = vlandev_find(ofproto, vlandev_ofp_port);
8384 return vsp->realdev_ofp_port;
8390 /* Given 'flow', a flow representing a packet received on 'ofproto', checks
8391 * whether 'flow->in_port' represents a Linux VLAN device. If so, changes
8392 * 'flow->in_port' to the "real" device backing the VLAN device, sets
8393 * 'flow->vlan_tci' to the VLAN VID, and returns true. Otherwise (which is
8394 * always the case unless VLAN splinters are enabled), returns false without
8395 * making any changes. */
8397 vsp_adjust_flow(const struct ofproto_dpif *ofproto, struct flow *flow)
8402 realdev = vsp_vlandev_to_realdev(ofproto, flow->in_port, &vid);
8407 /* Cause the flow to be processed as if it came in on the real device with
8408 * the VLAN device's VLAN ID. */
8409 flow->in_port = realdev;
8410 flow->vlan_tci = htons((vid & VLAN_VID_MASK) | VLAN_CFI);
8415 vsp_remove(struct ofport_dpif *port)
8417 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8418 struct vlan_splinter *vsp;
8420 vsp = vlandev_find(ofproto, port->up.ofp_port);
8422 hmap_remove(&ofproto->vlandev_map, &vsp->vlandev_node);
8423 hmap_remove(&ofproto->realdev_vid_map, &vsp->realdev_vid_node);
8426 port->realdev_ofp_port = 0;
8428 VLOG_ERR("missing vlan device record");
8433 vsp_add(struct ofport_dpif *port, uint16_t realdev_ofp_port, int vid)
8435 struct ofproto_dpif *ofproto = ofproto_dpif_cast(port->up.ofproto);
8437 if (!vsp_vlandev_to_realdev(ofproto, port->up.ofp_port, NULL)
8438 && (vsp_realdev_to_vlandev(ofproto, realdev_ofp_port, htons(vid))
8439 == realdev_ofp_port)) {
8440 struct vlan_splinter *vsp;
8442 vsp = xmalloc(sizeof *vsp);
8443 hmap_insert(&ofproto->vlandev_map, &vsp->vlandev_node,
8444 hash_int(port->up.ofp_port, 0));
8445 hmap_insert(&ofproto->realdev_vid_map, &vsp->realdev_vid_node,
8446 hash_realdev_vid(realdev_ofp_port, vid));
8447 vsp->realdev_ofp_port = realdev_ofp_port;
8448 vsp->vlandev_ofp_port = port->up.ofp_port;
8451 port->realdev_ofp_port = realdev_ofp_port;
8453 VLOG_ERR("duplicate vlan device record");
8458 ofp_port_to_odp_port(const struct ofproto_dpif *ofproto, uint16_t ofp_port)
8460 const struct ofport_dpif *ofport = get_ofp_port(ofproto, ofp_port);
8461 return ofport ? ofport->odp_port : OVSP_NONE;
8464 static struct ofport_dpif *
8465 odp_port_to_ofport(const struct dpif_backer *backer, uint32_t odp_port)
8467 struct ofport_dpif *port;
8469 HMAP_FOR_EACH_IN_BUCKET (port, odp_port_node,
8470 hash_int(odp_port, 0),
8471 &backer->odp_to_ofport_map) {
8472 if (port->odp_port == odp_port) {
8481 odp_port_to_ofp_port(const struct ofproto_dpif *ofproto, uint32_t odp_port)
8483 struct ofport_dpif *port;
8485 port = odp_port_to_ofport(ofproto->backer, odp_port);
8486 if (port && &ofproto->up == port->up.ofproto) {
8487 return port->up.ofp_port;
8494 dpif_stats_update_hit_count(struct ofproto_dpif *ofproto, uint64_t delta)
8496 ofproto->n_hit += delta;
8499 const struct ofproto_class ofproto_dpif_class = {
8534 port_is_lacp_current,
8535 NULL, /* rule_choose_table */
8542 rule_modify_actions,
8553 get_stp_port_status,
8560 is_mirror_output_bundle,
8561 forward_bpdu_changed,
8562 set_mac_table_config,